If you're that ahead on your current work, scoping out new stuff or filling in the gaps in existing projects is always a net win. This shows initiative, which you seem to have plenty of, which can only help your rating.

Of course there is the potential plot twist: your manager went on vacation purposely to see what you'd do in their absence.


Software Engineer

Elliot Kang

I feel your pain. You're in the dilemma of

1. stay in this weaker company and just practice good behaviors
2. go to a company with a much better engineering culture

Seems like you're choosing option (1) but even that is starting to get hard. The gap between the best and the rest is just so big. Here's a few options you can realistically take:

- Re-interview at FAANG (preferably Google/Meta) and accept the downlevel to L3, you'll still make more than Capital One
- Go to a later stage startup with a lot of ex-FAANG engineers. The issue with this is that there are a ton of washed up engineers and you basically get the bottom of the barrel (most of the time). But there are some good teams at startups
- Go to an early stage startup and into the crucible. 12+ hour days are common

FAANG will give you the best average outcome, the later stage startup can actually speed you up almost as much as the early stage startup (with 1% of the risk). The third option is 95% not worth it.

final note:

I'm not sure if non-stop shipping at an early startup is very good. As a new engineer you're going to be fighting against a bunch of engineers who have terrible SWE behaviors (one sentence Jira tickets, massive 1K+ line PRs, etc). Add that to the stress of a company going under and you're basically putting so much stress on yourself for no reason.

Yeah you need to build until you have a reliable resume generating engine. Go 80% building and then just do 1 Leetcode problem a day from something like Grind 75 (a list of DSA questions that are generally good prep, this is one of the harder lists):

<https://www.techinterviewhandbook.org/grind75/>

Also, to save time, if you're clearly not qualified for something (ie. you match 0 out of 12 job criteria), just don't apply. You save time and effort this way.

Hm. In the case that you're focusing on model training/serving, it is true that your skills will be much better compensated working for a hot AI mega-startup or Google/Meta. Almost all other companies simply don't have the scale/focus on AI to commercialize that skill. An alternative option is to try to work at a startup as a Research Engineer, but the startup will probably go under if they require a heavy R&D arm.

But isn't it an issue that ML skills have a limited employer set? Not really, it's niche but there are still enough companies that you'll always have a set of companies you can work for. The research topics will definitely evolve though.

Maybe you should think of it in tiers:

- Model building/serving at Google/Meta/OpenAI/Anthropic
- Model building/serving at a Tier 1-2 company (Databricks, etc)
- Research style engineering at a larger startup (Elevenlabs, Inflection AI, etc)

Just be careful with startups, a lot of founders have gone in way over their heads and just don't know it yet

I applied for a PM role once and they actually rejected me for the role but really liked my profile. The head of product basically carved out a MBA intern-like 6 month position for me. Ultimately they didn't have enough headcount, but that stuff does happen. Rare but it
happens

Ah yes. A master's from T5 will definitely help.

The problem with a majority of Master's degrees is that they are mainly course-only and don't do any focused research (this probably isn't what happened in your case though)

I don't think any degrees other than an undergraduate or PhD in Math/CS from a top school give any signal. Reason being, you won't get a rigorous enough education if your major is too easy. If you need more education purely for the sake of feeling competent or signaling to employers that you're competent, a graduate certificate or even a Master's/PhD is most likely not going to do it.

A better self-directed path is to either:

- do open-source for a medium-size project that's getting some traction (scope is restricted but your work is public)
- build in public and do side projects (your Linkedin posts + your project needs to be very high quality though)

Funnily enough, I dropped out of my Master's in CS and just started doing open-source. Open-source actually got me way more interviews than any piece of paper. My peers were DMing me many months later that they had zero interviews since they graduated.

Yeah just stack contributions on a mid-size project in the Pytorch ecosystem. Pytorch itself may be a little too big + hard to merge into within a few weeks. Maybe try a smaller one like Rahul suggested. `torchtune` comes to mind

I remember looking at my bank account while I was on my first job search, the balance was really low. This was in Fall 2023, right after the huge tech layoffs. My back was against the wall financially and the tech market was at an all time low (post FAANG layoffs). I had even dropped out of grad school. I picked up a part time job at a restaurant while interviewing. It was humbling, but it only motivated me even more to succeed.

Things seem rough right now, but I think this can be a great growth opportunity for you. Unless you've tried all of the following,

- writing a killer resume
- making really good side projects and building in public on Linkedin
- going to hackathons to network with amazing engineers
- DMing startup founders and make them beg to interview you

you shouldn't give up.

From personal experience, it's 100% a mindset shift. A friend of mine gave me the same explanation you're giving for not getting interviews. I rewrote her resume and applied to jobs for her. The new resume got some callbacks. It's not the wand, it's the wizard.

Just write a better resume.

Just be a better wizard.

A 30-second way of checking if your site got indexed is to ask ChatGPT about your website. Everyone uses Bing/Google in their backend

Diet will definitely affect your energy levels throughout the day.

- Keeping yourself hydrated will keep you alert, your body does bad things when dehydrated

- If you're just feeling a bit drowsy, natural forms of sugar like dried mangos help. Or if your company is boujie, Harmless Harvest 😆

Working in office can be exhausting, but don't feel like you have to be dialed in the full 8 hours on pure coding work. The whole purpose of going to the office is to have faster/richer communication with coworkers.

I've tried supplements but some are just snake oil, ie nootropics. Although I've been drinking carrot juice lately and it's helped a ton with eye soreness in the mornings after long nights coding.

Ah. Adding a custom model will skyrocket the complexity of this project. Custom model means you're taking care of:
- training
- eval
-model serving

... all of which are extremely time-consuming to get right.

I'd just go with GPT 4o mini, the token costs should be negligible if you're careful.

Yeah you're on the right track.

Also, something that may help is writing your own estimate on how long some tasks will take. Then you can compare with how long the task actually took and what made the task longer. This way, you can really beef up your estimation skills for project timelines.

There's some Taro courses on handling internship dynamics too

### Story Time

I wrote some of the core code for LlamaIndexTS in Fall 2023. Volume of code wasn't very high, codebase was small. Someone was willing to take me under his wing and provide some guidance on navigating open-source. LlamaIndex was really small at the time and they completely blew up in popularity. I then made a Linkedin post that generated 10+ interviews within a week. I was so overwhelmed that I declined 8 of them.

Fixed the resume and got 2-3 interviews a week online via cold applying. My Leetcode skills were horrible, so I literally got steamrolled by DSA interviews for 3 months. Around March 2024, my interview machine kind of stalled. So I just started re-responding to the founders who reached out from my Linkedin post. In April, some founder on Linkedin said they were hiring and I DMed them. Signed the offer a week later and started the day after Mother's day.

Concurrently, I reviewed a book on LlamaIndex and I got 2 paragraphs of fame for free (with some part-time review work of course). My only regret is not taking the time to shout out my entire extended family.

### Reflection

My timing was perfect, almost nobody else was doing GenAI open-source, and the code quality bar was very low. Open-source is getting harder. It still works, but as with any market, any new alpha is arbitraged out eventually.

That being said, exceptional execution is always rewarded handsomely. I think the open-source courses here are a great head start. You don't have to wait for a Jedi master like I did.

\*\*It'll probably take 3-6 months to get any real momentum. 1-2 if you seriously put an aggressive full-time effort into it.

Hm. Instead of what to code, I think how we code will probably evolve.

I've somewhat gone hybrid AI-assisted coding. Basically I start out a session by manually writing code and then have the auto-complete adapt to my style and suggest improvements. Occasionally I'll open up the chat window and ask for a change, but if the AI can't do it after a few tries I'll just code it myself.

Technical depth is still super important in this case. Pure vibe coding won't even get you past intern-level code complexity.

Easy wins:

- boilerplate code
- fast variable rename/import changes with auto-completes
- highly contained logic like "write this 5 line map-filter function for me"
- if you know exactly what you want but AI can find it for you (ie. some poorly documented `pytest` hook)

And if it's regex or CSS, try to get the AI to handle it 😂

People just expect too much out of LLMs these days. Sure, Meta might have fumbled a little bit on this one, but they basically have infinite money. Just by staying in the game, Meta and Google are effectively just killing the competition. And by the competition, I mean hundreds of startups that could have been potentially building their own LLMs.

I mean, look at any of the smaller foundation model startups. They have insane talent but they just can't play the infinite game that Meta/Google are playing.

Ugh. I know the feeling. A particular codebase I was working on was basically just an iteratively vibe coded jumbo slop mess. Unfortunately it is much harder to stay calm and focused in messy codebases.

Something that helped me mentally was not letting the codebase chaos get to me. This takes some getting used to but the mental strain of the legacy codebase will get less and less over time. This also means that you are not adding to the mess. You're trying to make the codebase better one diff at a time.

Instead of shooting for perfection, why not just shoot for a little better? For example, if there are bunch of nested conditionals, simply collapsing the crazy logic is a major win in readability. This way, you also don't spend 4x the time you originally budgeted for a task.

Marketplaces are not easy. You need to fill both the supply and demand side to make it work. Having tried to get a campervan marketplace off the ground before, I truly think that marketplaces are the most complex and difficult business schemes to pull off.


*Supply*
In your case, you need to grab listings from other websites and overlay your own UI on top.

*Demand*
Since you now have lots of pages with listings on your website, you'll rank higher on SEO. Additionally, writing good content will help people find your website.

Hm, platforms for data-intensive applications is still incredibly vague (in the grand scheme of things). Video is different from text or image or logging data.

Maybe you can find something more specific that you can be world-class at. Steve Huynh has a great video on it:
https://youtu.be/0kMWWm3BnBQ?si=SS0L3ypYPd9if5zq

This may feel limiting to narrow your focus down, but truly focusing is the only way to become world-class at something. But doesn't focusing on one thing limit the number of jobs you can apply to? Yes, but if you focus on something specific like payments processing, you'll get lots of payments processing interviews. And once you have mastery in one very specific thing, you can quickly gain mastery in adjacent tech stacks.

For example, data consistency is a huge thing in payments processing. Data consistency also happens to be a major issue in distributed systems. You can transfer some of the mastery over so you're never truly at square one again.

You can always add complexity. Random Name Picker could have been "distributed Random Name Picker". Now the product is 10x harder to implement and 100x harder to debug.

The craft is really in lowering the complexity instead of increasing it. Chasing complexity just for the sake of it will just result in wasted effort.


Trees. I was working with some guide data (in the form of decision trees) from a vendor where the API response came as a raw list of nodes with IDs attached to them.

The goal was to do a LLM-based enrichment scheme by feeding in the leaf nodes to a language model and having it extract key facts. In this case, getting pathing information on the leaf node was extremely valuable. (think URLs like abc123.com/products/trucks/tires)

So I found a way to identify the root node and did a BFS traversal to save the "filepath" of leaf nodes. This made the LLM-based analysis much better. Didn't think much of it until the vendor asked a few months later about how I did it 😅

Haven't worked in big tech yet but yeah.

Other quick examples conceptually:
- you can build a graph at scale using Redis. Since graphs can be expressed as sparse matrices, and sparse matrices are implemented on computers using hashmaps, and Redis is a gigantic hashmap/KV store

- Never use set() in Python, you get nondeterministic results. Instead use the OrderedSet object

- Workflows are generally DAGs, ie Airflow or even LangGraph. The better you construct workflows as composable DAGs, the smoother these are to debug/extend

Is this from an agency or the actual recruiting team? If it's an agency, it's most likely a scam.

AI startups are in a weird spot right now. Given that it was easier (for a while) to raise $2 million in funding than to get any revenue, there are a ton of GPT wrappers that will likely implode in the next 12-18 months.

Why not just start your own company? If you have some good SWE experience and have started a company before, you're well-equipped to be a founder already. Founding engineer compensation is doing a lot of the work of a founder, except you pay a heavy "de-risking" penalty for not being the one securing funding. (You own 2% at most, instead of 20-30%)

This is a very opinionated take, but unless you see a tremendous personal or financial growth curve, it's generally a very bad deal.

It can be a mixed bag with first-time managers.
Personally,

what worked really well:

- Over-communicating. (at least in the beginning) Since they're new to the role, giving frequent updates sets them up to succeed. There are a lot more variables that come with managing up, down, and sideways.

- Telling them your goals. Your manager has the most say in what you'll be working on. But they can't support your goals when they don't know your goals.

what didn't always work well:

- Asking for high-level feedback. To be fair, this was partly my fault as I wasn't very specific. My questions were vague like "How can I improve my communication with the team?". Instead, pinpointing a specific example followed by "Here's how I currently add value during architecture discussions, how would you handle this?" elicits extremely specific feedback.

- Expecting your boss to have everything figured out. Since this work relationship is new, you're both trying to figure out the best way to work together. Sometimes they won't give actionable feedback. Like all relationships, it takes effort on both sides to make it work.

Ethan Evans has a great Taro course on this, wish I took it earlier. 

One way you can work on live code with friends is by going to a hackathon. There's a bunch in the SF Bay Area

Lots of topics here.

Try this:

Give yourself a hard stop from work at 6pm and go eat some dinner. This lets you recover that mental bandwidth

Do one thing at a time. Go front-back from the resume review to the interview process.

- Get your interview-generating machine churning out 3-5 interviews a week.
- Review technical concepts.
- Practice DSA.
- Practice System Design.

If you're stuck at step zero (no interviews),  get a laser focus on exactly what type of role you're looking for. This not only keeps you grounded, it gives you a very specific goal that makes the rest of your job search easier.

By knowing **exactly** what you want, you only need to prepare for 1 type of interview process. Then you only apply to 1 category of job, and that will net you wayyyy more interviews than an unfocused resume.

AI/ML Eng @ Series C startup

Wow your colleagues sound a lot like me when I was doing research in my Master's! My code was complete garbage. They most likely just don't know any better.

Some possible action steps:

- If it's a monorepo or not that big, you can introduce a linter with custom rules
- Offer constructive feedback. Their code works, but it's ugly. The polite way to say this is "Code works great, just a few style changes" on the PR reviews.
- Find people who are either underperforming due to the codebase or are tired of dealing with shitty code and work with them. **Suffering people are the most open to change**.

This does require some tact though. People don't like their baby being called ugly. Depending on how serious this problem is for everyone's productivity, the harder you can push.

Yeah inbound is inbound. On the flip side, I've had dozens of founders reach out and ask if I wanted to interview with them.

Generally, if the recruiter is from the company, that's a free interview. Otherwise, the inbound doesn't really count. Recruiters are basically salespeople for companies and most of them just don't do basic checks like tech stack.

Good content can get you really good inbound, you just need to invest the time to do so. Given how everyone's busy with their day jobs, the bar for content is not that high. It's about being consistent and writing meaningful stuff (which I too only do on a sporadic basis haha)

This course is an absolute gem, I got \~300K views on a post by following this:\
<https://www.jointaro.com/lesson/tmSAAWJjLcLWqeGWOl4t/case-study-how-i-get-500000-linkedin-post-views-per-week/>

It might actually be better that you work an hourly role right now, at least in the short term. Since you're getting back into the tech industry, you can use your free time to find a great role that you actually like. Letting your muscle memory (with work stuff) kick in while the stakes are lower is actually a blessing.

You can also just start a side hustle. I met a guy who had a day job as a Data Scientist and ran a $50 million HFT fund in his free time.

By the way, your level should probably even out within a year even with some "lower level" short-term gig between jobs.

Agreed. Getting an offer and then getting it rescinded the following spring just wastes everyone's time

I have a pretty bearish take.

### Tech Layoffs

This is probably the new normal for the tech industry. Its gone back to being ultra competitive, as it should be. $1M salaries don't come for free.

Eventually, this will just shake out all the developers who only thrive when the markets are good.

### Job Searching

Branding oneself solely based on big company prestige will not help anyone. Why? Because ten thousand people branding themselves as "mid-level ex-FAANG SWE" are competing for the same jobs.

They'll need something extra, like side projects or open-source.

### AI Coding Tools/Productivity

Eh. The issue with AI coding tools is that they're appealing to two distributions of SWEs:

- mediocre engineers who just need really basic things done 
- amazingly productive engineers who can prompt the LLM to do exactly what they want

Mediocre engineers don't know the difference between quality and slop. 

The top SWEs know what they want at such a granular level that the LLM has all the necessary context to output the right answer.

So eventually we'll end up with the mediocre engineers staying mediocre and the 10x engineers will become 100x.

### Overall

Tech is becoming hyper-competitive and it's going to be really hard for people to compete. But if you're top-tier, then you'll make even more money.

I went to Japan in January and there were a lot of American startup founders working at the local Share Lounge (Shibuya, Naka-meguro). I heard `shadcn` so many times that I thought I was back in San Francisco.

Basically, after a few years in tech and gathering experience/cash, you can get different types of jobs too:

- work at a more stable Silicon Valley company (Oracle, Cisco, Intel, etc)
- opt for the digital nomad life with remote job (COL is very cheap overseas)
- make your own startup in a foreign country like Japan (costs $1K/month in Tokyo)

If you're just going with the flow in these 1:1s, you may be wasting time.

Some possible questions: 

- What's taking up the most time for you when it comes to these tickets? 
- How well has your boss been keeping up with big picture goals/how could you help them achieve those goals? 
- What are some things that are unproductive for you that you spend too much time on?

These are all basic discovery questions coming from the mindset of "how can I make my life/boss's life easier?"

Not all of the conversation needs to be ultra-structured. I try to get the important things out of the way first (productivity, team dynamics, workflows, etc) so that the later half of the 1:1 can happen more organically.

ah. there are probably places too, just threw out the most ridiculous example 😭

From what I've (briefly) seen with larger systems so far, theory and practice are so so different. Why not build something like a distributed text editor and stress-test it against a bunch of cases? Hopefully you can do this in addition to a course or self-directed reading. But reading alone is nowhere near enough to be proficient at something.

**Moving Companies**

Is there a way you could stay despite having an unpleasant manager? Hopefully you can grow your scope to more than what your manager assigns to you. If you like the company but just have an issue with one person, I don't think that's a valid reason to move.

**Growth**

There's other ways to grow as an engineer than just your day job. You can get involved in open source, do side projects, or just hack with other engineers on the weekends. This only applies if you have really good WLB and can invest time into this of course.

That being said, you should definitely find some form of mentorship whether its Taro or a skip level mentor.

Where are you located? If you're in the Dublin/Pleasanton area, it's not that hard to find a referral. Try asking around at the Bay Club Pleasanton or just hang out and talk to people at Beer Baron on Thursday nights.

Basically, there are a ton of Workday employees in the area and you just need to get out there. I live in Livermore and our version of Workday is the Lawrence Livermore National Lab.

In your case, I think open source is way better. You can just focus on dev and not have to care about the virality aspect, as the creators of the library are responsible for that.

If you work on a side project, then you still have to do some of the legwork to make the project noticeable to recruiters.

In general, I think open source is actually way better for people with the goal of getting noticed... as long as you pick the right projects.

In the long run (over several years), self-directed side projects will make you a better engineer probably.

Doing a Master's and then getting H1-B visa is getting harder and harder every year. The visa lottery is difficult enough that even O-1 is starting to get hard as well. I'd go with trying to transfer within Microsoft, as you're already working there. 

It makes no sense to quit, do an expensive Master's in the US (100 Lakhs or more), and then try to rejoin a top tier company to get visa sponsorship.

Just apply. Let the company decide leveling at the offer stage. But if you're not applying, you have no chance anyways.

"You miss 100% of the shots you don't take"
- Wayne Gretzky

There is no apples to apples comparison, FAANG just gives a ridiculous guaranteed amount of money... no matter how many different ways you fudge the numbers.

TC doesn't mean jack in startup world. You're always taking an outsized risk by joining a startup that may/may not IPO. By default, all non-public company shares are inflated, since there's no liquidity in them. Cash salary is the only part of TC you can depend on there.

Think of it this way: joining FAANG is like investing in the S&P 500, you'll get a decent return for the effort/time you put in.

Joining a startup is like investing in Gamestop. It could go to the moon or go boom.

i work 100% remote and have less than 1 YOE in software engineering

Engineering team was a similar size at my last job, seed stage. The general expectation was 3-4 months.

For mocks, you can use Figma if you need super high fidelity mocks, but that AI tool Alex mentioned should be enough.

For design docs, I found that either Notion or Google Docs in pageless format works really well.\
\
This gem should be a super-course:\
<https://www.jointaro.com/course/frontend-system-design-masterclass-building-playlists/>\
\
In the `Defining Requirements` section, there's a link to a design doc for a 1-week long feature build from Alex.

Something a mentor told me is that even if you're looking for a new job, you're still getting paid to work at a minimum 9-5.

idk if this helps, but it helped me a ton

It's a skill issue. No matter how "bad" the economy gets, there are going to be people who land tons of interviews. Your job is to be one of those people.

You can still land interviews as a new grad. I even dropped out of my Master's in CS and landed 3-5 interviews a week in Fall 2023, when the economy was supposedly terrible.

There will always be a market for exceptional talent. So show that you're exceptional.

As they say at Taro:
- get good
- land interviews
- pass interviews

MS doesn't count as 2 YOE in Big Tech/FAANG. People don't count it at startups either.

I would just apply anyways. The "shot" only counts if you make it to the interview loop.

Heck even at seed stage... most of the time, the existing code wouldn't have handled all the common cases. Then when you say "Refactor this code for me", Claude slashes up your code even more.

Conventional wisdom just doesn't work. You'll have to decide for yourself whether you want the accelerated growth of a startup life or not.

Big tech **on average** produces better results. But you are not the average (or so I hope). Remember these folks did really really well at FAANG compared to the average actually. They're telling you what worked for them.

From the information in your post, I can smell you've already made up your mind about the startup. Just do it.

Having grown up in the Bay Area, San Jose **does not equal** SF. You'll be driving up 1 hour to SF, which is doable but very painful.

**Career**

The AI appeal is really nice, but you'll seriously have to hustle to even have a chance. The competition is pretty intense. Startup scene is cool but it can get old when people are talking about Claude in a nightclub at 1am.

AI events are fun, there's a ton going on every day. They're okay for networking with fellow engineers. Founders usually speak at these events or work out of Shack15. It's not too hard to get a referral to Shack as of now, but that could change if they initiate a waitlist.

**Social Life**

I've been in a long-term relationship for 5 years, but my single friends tell me dating life is atrocious in SF/San Jose. The colloquial term for San Jose is "Man Jose".

Just leave things as is. You don't benefit at all from disclosing more information. Water under the bridge

Panicking in this situation is not going to do anything. Whenever you feel overwhelmed, step away from the computer and take a deep breath.

Something that concerns me is that you're instinctively falling into work anti-patterns.

1. Multi-tasking - Line up all your tasks in order of priority. Break down each task down so much that even a LLM could code up each atomic mini-task. Block 4-8 hour blocks of time to only do coding. Do the tasks one by one. Context switching is the ultimate killer of productivity.

2. Ask for Help - If your teammates are saying that you ask questions about trivial things, ask them the deeper question of how you can find these things on your own before asking next time. If they don't want to provide that advice, make sure to spend 15-20 minutes finding your own solution *and then* ask for help.

3. Work Longer Hours - Working longer hours only helps (barely) boost what's already working. In your situation, nothing's actually working so you need to revisit your work habits before you even do this.\
   \
   Go through Alex's course on "Junior to Mid Level Engineer" course. It singlehandedly helped me turn around my performance at my previous job. I know it seems like the world is crashing down, but sometimes you need to regroup and really make a strategy for moving forward.

<https://www.jointaro.com/course/grow-from-junior-to-mid-level-engineer-l3-to-l4/how-this-course-works/>

Morgan Wallen. There's nothing more encouraging than some slow, sad country music.

Don't have enough context to answer. What's the size of the company? Is this at a tech-first company or a company where tech is a support function?

Otherwise, we don't know the severity of several senior engineers leaving.

There's some good content here:\
\
<https://www.jointaro.com/question/YqrpDDCPbsGpg0lwrKXY/looking-to-transition-to-ml-engineer-what-do-you-want-to-know/>

I can't pick your niche for you, but here are some good constraints I wish I knew. Pick something that

- won't require you to make a top-tier ML conference paper
- has a mid-size open-source library (5-15k Github stars) that you can be a Top 3 contributor in within 3-6 months
- is actually interesting to you. You don't need to already be a user, but you should have an actual interest in playing around with adjacent tools

Hope this saves you time

\*\*I wrote this in case you pursue ML. Although you did mention "data roles"

As a software engineer, you have the power to will things into existence (with some additional typing of course). You don't need permission to build anything.

PM is very often a trap for software engineers, and it's really hard to go back to SWE once you've moved to the dark side. As Rahul hinted at, a lot of big tech PMs basically just push metrics and fill out Jira tickets.

Why not just focus on being really, really good? Opportunities to collaborate with exceptional people will open up as you do.

The lightsaber is only as strong as the wielder.

Just turn off all AI tools and use Google, read the docs yourself, and build stuff.

AI hallucinates convincingly, so it's hard to actually increase your productivity with AI unless you take a nuanced approach. You still need to do your homework on specifics.

You're not alone. I was jamming my keyboard begging ChatGPT to build me a multi-layered Dockerfile and it just kept spitting garbage 🫠

Navigating a somewhat adjacent situation myself, you're not alone 😓

now that i think about it, my answer was very un-nuanced. you can ping me on linkedin if you have specific questions, I ran the largest AI meetup in the bay area for a year

Try cerebralvalley.ai


- hackathons at AGI House in Hillsborough (not the SF one) are pretty fun, top tier engineers show up. Andrew Ng was squinting really hard at my failed demo one time 😭

not searching rn. but yeah. I found that all mainstream social media has garbage sponsored content as 90% of the feed, making something with less sponsored stuff is awesome

Tbh, you're probably not making progress on your DSA/interview prep because you live in an attic and have no coworkers to talk to. Your environment outside of work often affects your work deeply. So maybe instead of $400K being your immediate goal, just get out of your parents' place.

Personal experience: I dropped out of my Master's and lived at home for a year before finding a SWE job. It was a terrible existence. I still do live at home but have a job now so it's much more bearable.


Sorry don't have any inside info. But you can find a lot of this info out yourself if you can secure the interview.

The best way to find out is to apply now and secure the interview. Then you can ask the hiring team what the interview process looks like + some detailed follow-up questions. If you're underprepared, respond slower to Elastic and push back the interviews so you can prepare

This is gonna ruffle some feathers, but I think that the best way to prep for technical interviews is just to do more technical interviews. The nerves are the worst part for most people. You can farm early/mid stage startups for technical interviews tbh (if you already generate tons of interviews)

Sounds like an awesome idea. Even better, if you find any alumni from your school's student clubs that you are involved in, that can make the connection even better

Just leave. There's plenty of other AI startups with funding. It looks like your team couldn't find customers before the lights went out. That's too bad. Dust yourself off and go find something else to do

Whenever someone asks you if everything is clear, just pause for a few seconds and think through whether you can visualize the implementation clearly (lines of code). The silence may be awkward, but it'll save you lots of headaches later.

^ what Rahul said.

Now that I think about it, this is also how you proactively get good at contributing to open-source.

There's some great input here. So why does school matter? (at least initially)

People are always looking for signs of exceptional ability. And at the beginning of your journey, school is often the only indicator of ability if you have zero internships or cool projects.

Top company is an even better filter than school because it's a more recent indicator of ability. Even at top schools, very few people actually end up succeeding as much as you think (FAANG, IB, etc)

Ah I see. Then yeah maybe it would be better to work under someone with much more experience for a bit. You can always calibrate your work-life balance up or down as you progress.

I'm also on the extreme WLB end right now, went on 3 trips in the past 2 months aha

If it's generative AI, I'd wait until they're slightly later like Series B or C. There's so many knockoff GenAI companies that they're extremely unpredictable right now. 

YC-backed doesn't mean anything these days, as YC is now shovelling out AI startups like there's no tomorrow. If you don't plan to stay longer than a year, your shares won't vest anyway so I would just go somewhere else with the plan to stay at least 1-2 years.

I wouldn't trade the free time, as you could be sharpening your skills on your own time and not someone else's. That said, it requires more discipline than when you're a busy employee.

and you can probably do that in the "tell me about yourself/background" blah blah blah

Now that I'm re-reading this, "cleaver of truth" sounds really corny. The point is, you need to proactively address any concerns your interviewer may have before they bring it up 🫣

It's a trading firm. Compensation should not be a taboo subject. I understand that many people try to dance around touchy subjects, but that only raises more alarm bells in the interviewer's mind. And compensation is only a lagging indicator of organizational problems, so feel free to add 1 or 2 more of your complaints when people ask. (gracefully, of course 😅)

When someone has doubts about you, your responsibility is to cut through that doubt with the cleaver of truth.

I started interviewing for new roles 4 months into my previous job. The first thing interviewers asked was why I was leaving so early. So when I noticed the trend, I pre-emptively told them why I was leaving my current job (lots of turnover, etc). They never asked again.

Everyone has to play the interview game. But you get to decide how you'll play it. Are you going to dodge all the real questions until you get eliminated or are you going to proactively attack the interviews and get the damn offer?

That's up to you

Not informed enough to know whether this works for higher levels, but this is a good strategy for now at least. Erring on the side of overcommunication is better, especially if you're fully/partially remote.

Don't be discouraged if your boss doesn't respond, he/she will still read those updates and have some piece of mind. Plus if you're getting blocked on a 5-day task, you'll be able to catch it Tuesday and not the following Monday

Oh. and. Try to carve out 4-6 hour blocks of time where you're doing some deep work coding. Alex mentions this in his ["Maximize your productivity"](https://www.jointaro.com/course/maximize-your-productivity-as-a-software-engineer/) course, and that really helps from personal experience

Like Sai and Charlie mentioned, it takes a certain amount of effort to make AI actually work for you when coding. AI is good for:

- extremely common code patterns
- shuffling simple stuff around (variable name changes, small function refactors)

Larger design decisions that span multiple functions/scripts don't work very well. Similar to how Alex mentions breaking down units of work, breaking down work even more with atomic tasks helps you get a way better ROI on AI tools. I use this process:

- break down tasks to the function level
- insert some comments to annotate what I want Cursor to do
- use the chat interface and feed in context for each mini-task
- audit your code by looking at it/testing it

Rinse and repeat.

The way I like to think about it is that AI is 100% useful on 20% of your work as a SWE (doing mundane stuff like regex, etc). Cursor can't help you distill requirements or make a great design. Cursor also won't help you manage expectations with your boss.

To add onto this: just don't do any of this work on a company laptop if you plan to commercialize

If you have the flexibility to, travel. I'm on my third trip in 3 months and it's been so helpful in resetting + building new work habits.

It can be difficult to break out of bad habits sometimes, so physically relocating even for a few days can make your brain flow differently

Both work. However you need to answer this question first: why should this person hire you out of hundreds, probably thousands of applicants?

**Why** is so so important. And the way that your "why" manifests is by doing deep work on relevant projects to the CS subfield you want to be in. It took me the longest time to put this idea together mentally

Hm. While your intentions are good, I think you're leading too much with "look at me I found a problem in your system". Could you try just reaching out with the premise of wanting to learn more about the company/team? Then you can lean more into "I built something and found out some cool stuff about Netflix' production app this way" as the conversation warms up.

Open-source really is king when it comes to this kind of "attention grabbing". Side projects can be hit-or-miss unless you're cracked.

Awesome that you're switching to SWE! Piggybacking off of Alex:

- For non-frontend roles, I don't think they'll ask you too much framework-specific questions
- Leaning into your strengths as someone who works with data-intensive applications will help a ton
- As Alex said, in practice TDD can get in the way of innovation. You can just follow your new team's testing culture when you join

Finally, on knocking down those doors to get those interviews:
- If your project is somewhat similar to what the team is looking for, you can ping the hiring manager, or on the application submit your project code as a work sample. I've gotten several interviews this way, and also gotten people who were on the fence about me to at least move me to the next round of interviews

Like Rahul and Charlie said, I think it's a good sign that you're still hanging in there. All things considered, your case is on the better side given all the layoffs/turbulence over the past several years.

I graduated college in 2019 and some of my peers never recovered

no, this is just a measure of how big of an AGI House fanboy/girl you are 😆

You have no idea how much pain running a startup can bring. It really can be like having broken glass for breakfast every single morning.

I think the premise of your questions should be different. Life is a marathon. Instead of jumping the gun and trying to learn everything at once, why not learn aggressively at a sustainable rate? Most of the time, the way to do this is to get a hybrid/on-site SWE job and moonlight on the side. Then, you can go 100% remote for even more freedom. Then, you can quit your job once the startup makes $10-20k USD/month.

I know you have this patience: you stuck around for a CS degree. But going 5 years at full speed doesn't do you any good if it requires 2 years of therapy to recover.

To add to Alex's statement, it's easy to misread data. For example, those metrics like MMLU don't mean much when you're trying to make these "AI agents".

Benchmarks, by design, look really good. But AI projects/papers often solve a really specific problem, so blindly hoping research results generalize to your use case doesn't work.

When learning any science, the real skill is finding which information is relevant/actionable to you. It's 99.9% noise out there in GenAI. You need to do side projects/implement papers yourself to get any real alpha in GenAI tbh.

I'm also curious about this question. Currently, I'm writing some Terraform/Chef to spin up resources to run distributed training jobs on the cloud.

We don't all have a 10000 GPU megacluster to run pretraining on 😿

ooh. I'd take the Anduril SWE option. Not because your stocks will hit the moon, but because you'll stay a software engineer. 

As a business engineer, you wouldn't do as much hardcore coding.

**take this with a grain of salt though, some more informed people will have better opinions. But one thing is for sure, please don't count on the non-cash portion of TC at a non-public company (you'll only end up disappointed)

Low. Kaggle ranking is just a number. Open-source and building in public generates far more inbound than Kaggle. You can generate dozens of inbound interviews this way. Just do cool stuff, then write a little about it. Rinse and repeat.

The reason for this is **people want to see cool shit**. A ranking number is boring.

To answer all your immediate questions:
Papers are a good way, but they take lots of time and your work is unlikely to be rewarded. There is no certainty that you'll publish a decent paper, if at all. Generally ML people don't care about applied ML papers as much. Worthwhile top tier papers take 12-18 months from idea inception to publication acceptance/release **at a minimum**. Second tier conferences like AAAI and KD are still good, but less prestigious. Workshop papers are in much lower regard to real papers.

Some general points:
- ROI on papers is low, given how much time/effort you're putting in. If you want to do purely model research though, this might be the only way
- Very few ML engineers are purely doing model work. The majority are doing systems for ML, SWE for ML, or infra for ML.
- open-source is pretty good ROI if you know what you're doing, there's a Taro course on this
- Switching from engineering work to science work has it's own risks. It's hard to measure research productivity and you might find yourself at a C-tier research shop with some bums

The main point is to always line something up as you never want to be in that OPT phase. You can do this! 💪

Alex's advice is spot on. Being proactive/working well with others is a good defense in itself. There are ways to be proactive about your fate without playing full-on offense, such as covering all your bases like Alex mentioned

You're definitely not alone in disliking this manager person. And you're not the crazy one (in case anyone tries to gaslight you)

Given your visa situation, I think it'd be good to stick it out for some time at Amazon. Your number one goal should be to buy time as someone on a work visa. It's like playing poker or trading stocks: never get completely wiped out. If your visa expires, it'll be way harder to find another job in the states.

IMO visa >>>> compensation right now. If you find another company willing to sponsor you though,  then yes the visa situation is taken care of and then you can get greedy on compensation.

**In case others have further input, how bad is the compensation difference? I think it's okay if your comp is lower than the "average" people on Levels.fyi

Hm. I graduated at the same time too. With all the data points you've given, I don't think this is even a tech stack issue. You're **using bootcamps/degrees as an intellectual crutch**.

Like,

- more bootcamps/degrees won't help you feel smarter
- switching tech stacks won't help you build cooler things

You still need the core SWE job search things like writing a great resume and passing the interview rounds. But doing the same thing everyone else has been doing... phoning in for a 9-5 isn't going to cut it anymore.

If I were you, I would

- Find a meaningful [open-source](https://www.jointaro.com/course/become-an-open-source-master/) project that you'd be willing to commit to every single day for \~2 months
- Write a killer resume. If you're not getting any callbacks after 15-20 applications, it's a dud. **Every line should be a banger**.
- Practice interviewing, once you start landing at least 2-3 a week. Your resume isn't an effective hook until then.

You can do it. If you need some help, we can set up a much more granular action plan to get this done. You're in a much better situation than I was in mid 2023 (MS dropout, no CS degree). Ping me on Slack or Linkedin if you need to

### Personal Anecdote

You're in a bit of a hole, but that's fine. I've been in a similar situation before. It's never too late to turn it around. My brutally-honest resume is:

- Graduated 2019, bombed maybe 30-40 interviews (including Confluent when they were $1B valuation)
- 2020, got wrecked by pandemic + worked some odd jobs
- 2021 - 2022, played mobile games on my iPad + did CS courses
- 2021-2023, went through the motions doing a Master's in CS
- 2023, dropped out and did open-source
- 2024, finally got an AI job after I got tired of working in a restaurant (tips were good though)

Not the optimal path. But hey, at least I have fun building things now. And I washed dishes really fast after Thanksgiving dinner.

lol. So one of my friends had a similar interview experience when he did his onsite at Amazon. It was two questions and they just chatted the rest of the time. He got an offer and ended up being a really high performer at Amazon.

But back to your situation. The question behind your initial question is whether your super short onsite is a sign you failed the interview.

You're overthinking after the interview. You already did everything you possibly could do. They could have put you through a grueling interview loop if they wanted to. **If they wanted to, they would**.

TLDR; you're overthinking, just wait it out.

I don't think you should have any guilt regarding "I'm getting paid more now, therefore I should work more hours".

#### Getting Things Done on Time

That being said, your projects need to get (mostly) done on time. Maybe you could discuss with your manager why your projects are running behind. A couple common issues are:

1. Scope creep. A 3-day issue turned into a 7-day issue because you or someone else added to it. **Fix**: Make a follow-up ticket/diff for the new additions.
2. Getting rug-pulled. The team agreed to the scope of a ticket/project but they say they wanted way more features at the finish line. **Fix**: Get people to agree to written requirements before implementation.
3. Code review takes too long. Your code is too annoying to review, so nobody does it. **Fix**: Alex discusses splitting diffs in the [Level Up Your Code Quality](https://www.jointaro.com/course/level-up-your-code-quality-as-a-software-engineer/) course

Don't be afraid to call people out on (1) and (2), as it's now your reputation on the line.

#### Gracefully Handling Lateness

Regarding being late for projects, things always come up in a startup. By default, some projects are going to be late. It's how you handle the lateness I think, that really matters. Some key things to ask yourself are:

- Are you communicating with the team that solving issue A will delay project B by # days? How well is news like this received?
- Is your dev pace matching/exceeding current members of the team?
- Who is doing your performance review?

#### Performance Shouldn't Be a Surprise

Finally, your performance review is basically at the mercy of whoever is in charge of that. I think that given you're not even 90 days in, it's impossible to give a definitive or close-to-precise answer. If someone got let go in 90 days, that means they legitimately did nothing, so I wouldn't worry too much about that. (if that's only a single data point)

To answer your question, **there is never a guarantee that your team will let you know that you're underperforming**. If I were in your shoes, I'd bring up your concerns with your manager (who I'm guessing will do your performance review) and see where you guys can iron things out. If possible, it's good to gather your thoughts and do this asap.

Hi, seems like you're on the right path! I understand your urgency, as it's a 6-month contract and you're trying to secure a full-time gig.

I found these courses really helpful at my last job where I had to deal with some serious leadership drama, switch teams, and turn around my performance as a "Research Engineer"-turned "SWE with a sprinkle of AI".

### Courses

#### Junior to Mid Level Engineer

#### <https://www.jointaro.com/course/grow-from-junior-to-mid-level-engineer-l3-to-l4/>

Helps with code quality, communication, and distilling requirements for projects.

#### Leveling up Code Quality

#### <https://www.jointaro.com/course/level-up-your-code-quality-as-a-software-engineer/>

Really helps you get into the feedback loop of improving code quality with the more senior engineers, as they'll get "setting the bar" for what gets pushed into the codebase and what's rejected.

#### Performance Optimizations, for work

#### <https://www.jointaro.com/course/maximize-your-productivity-as-a-software-engineer/>

If you feel like you could be way more productive at work, you're not wrong. It's an iterative process, but there's some good pointers here. Alex also politely tells you not to be a lazy bum.

\*\*Doing these courses in sequence yielded really good results. It takes a few days to go through all of these, and maybe 2-3 months to really internalize.

### Masterclasses

#### Shipping Real Features

#### <https://www.jointaro.com/course/frontend-system-design-masterclass-building-playlists/>

I borrowed the design doc format for this masterclass and got amazing feedback from my team. There's three ways to use this IMO:

If you're at a larger company, hopefully they already have a uniform design doc format and you can just take away some nuggets of wisdom from this class. If you're at a startup, chances are you aren't getting much guidance, so you'll have to bootstrap a format on your own and iterate on it. You can also just practice by **writing design docs for your side projects**. Here's a somewhat incomplete design doc I wrote for making a [Figma AI copilot](https://docs.google.com/document/d/16hKvPpoF2Q8fyJd1qnvhHtuRPMJKkfiYIiWvmBcTC4s/edit?usp=sharing)

These can be fun to write in your free time + you get to really think through your ideas.

Anyway, hope this helps.

Wow you write well! I really like your point about being proactive finding work to do. This definitely makes life easier, as you're no longer at the mercy of your Jira board

In my enthusiasm, it looks like I missed a few points 😅 Some things I notice about really, really good engineers is that

#### Breaking Down Work
They're just so good at breaking things into atomic tasks. If there's a whole new workflow, they break it down into as many trivial tasks as possible and knock them out.

Some of Alex's videos on "how real system design/feature planning is done" are super helpful. There's like 3-5 of these at least somewhere.

#### Getting the right people involved
If there's a feature that involves approval from 3+ people, I've seen them just get ahead of the work and gather requirements super efficiently.

TLDR; good engineers are really good at **doing the preparation before coding** so they have to write 70% less code when the implementation actually begins. WLB does exist, but I think you'll have to fight hard for it.(being efficient, constantly improving, picking the right companies/teams)