Paper Reading: MyRocks: LSM-Tree Database Storage Engine Serving Facebook's Social Graph

Session led by [Dharin Shah](https://www.linkedin.com/in/dharinshah/)

**Event description**

Facebook uses MySQL to manage tens of petabytes of data in its main database named the User Database (UDB). UDB serves social activities such as likes, comments, and shares. In the past, Facebook used InnoDB, a B+Tree based storage engine as the backend. The challenge was to find an index structure using less space and write amplification.

LSM-tree has the potential to greatly improve these two bottlenecks. RocksDB, an LSM tree-based key/value store was already widely used in variety of applications but had a very low-level key-value interface. To overcome these limitations, MyRocks, a new MySQL storage engine, was built on top of RocksDB by adding relational capabilities. With MyRocks, using the RocksDB API, significant efficiency gains were achieved while still benefiting from all the MySQL features and tools. The transition was mostly transparent to client applications.

The paper was published at the **International Conference on Very Large Databases (VLDB)**: [**https://research.facebook.com/publications/myrocks-lsm-tree-database-storage-engine-serving-facebooks-social-graph/**](https://research.facebook.com/publications/myrocks-lsm-tree-database-storage-engine-serving-facebooks-social-graph/).

Paper authors:

- [**Yoshinori Matsunobu**](https://www.linkedin.com/in/yoshinorimatsunobu/)
- [**Siying Dong**](https://www.linkedin.com/in/siyingdong/)
- Herman Lee

Often Forgotten Team Considerations

We wrap up with two often-overlooked but highly consequential factors when selecting a team: on-call health and code review speed.

Here are the core points from the lesson:

- We explain that on-call quality can make or break your work-life balance, and many early-career engineers don’t evaluate it; the two key questions are how *frequently* you go on call and how *painful* it is (night pings, fire drills, number of incidents).
- We highlight that not all on-call is equal—middleware teams or teams responsible for triaging other teams’ outages often have disproportionately stressful rotations unless strong runbooks exist.
- We emphasize that code review speed is critical for learning and momentum: slow iterations (1–3 days per review) compound quickly, turning a small change into a week- or multiweek-long process and severely limiting growth for junior and mid-level engineers.
- We define healthy expectations: reviews returned within a day are ideal, &gt;1 day is a yellow flag, and &gt;2 days per revision is a red flag because it throttles how often you ship and how fast you improve.

Paper Reading: Beyond Next-Token Prediction: Enhancing Language Models with Multi-Token Outputs

Session led by: [Lucia Mocz](https://www.linkedin.com/in/lucia-mocz-ph-d-35a98a1b6/)

In this talk, we will look at groundbreaking research on advancing the training of large language models. Traditional models like GPT and Llama rely on next-token prediction, which can be expensive and time-consuming, but this research suggests a more efficient approach: multi-token prediction.

**Key Highlights:**

- **Enhanced Sample Efficiency**: Training models to predict multiple future tokens at once increases sample efficiency, leading to significant improvements in downstream capabilities.
- **Innovative Architecture**: By employing n independent output heads on a shared model trunk, better performance with no additional training time can be achieved.
- **Superior Performance**: 13B parameter models outperform strong baselines on generative benchmarks, solving 12% more problems on HumanEval and 17% more on MBPP.
- **Algorithmic Reasoning**: Multi-token prediction fosters the development of induction heads and enhances algorithmic reasoning capabilities.
- **Faster Inference**: Models trained with 4-token prediction are up to 3 times faster during inference, even with large batch sizes.

Join us to explore how multi-token prediction can revolutionize language model training and performance, especially for coding and natural language tasks.

Don't miss this opportunity to learn about the future of language model training and its practical applications!\
\
Paper: [**https://arxiv.org/pdf/2404.19737**](https://arxiv.org/pdf/2404.19737)

How To Do Research

We walk through how to actually evaluate companies and teams, emphasizing that good information is hard to find and rarely lives in public reviews.

Here are the core points from the lesson:

- We stress that big companies aren’t monolithic—cultures vary widely across thousands of teams—so the most reliable way to assess a company or team is to tap into our network, where trusted people can share accurate, firsthand experiences.
- We highlight why building deep relationships in school and at work is invaluable: our network expands our access to real data over time, and meaningful connections (teammates, managers, peers) often shape our opportunities more than technical ability alone.
- We suggest complementary methods—reverse-interviewing during the hiring process, reading engineering blogs, following tech news, and checking curated communities—to gather signal, while warning that Glassdoor skews too positive and Blind skews too negative due to self-selection.
- We show that even venture capital, which allocates billions in tech, runs almost entirely on networks and backchannel references—underscoring that our own career decisions benefit from the same relationship-driven information flow.

Engineering Manager Misconceptions

If you want direct interview support from Alex and Uzair, take their live Maven course with discount code **TARO10**: <https://maven.com/alex-kliotzkin/acing-em-interviews>

This segment dispels a common myth in engineering leadership: that an engineering manager must be the most technically knowledgeable person on the team. Instead, it reframes technical leadership as an exercise in empowerment and strategic growth—both for the team and for the manager.

- **Technical mastery is not a requirement for effective engineering management.** The core responsibility is ensuring delivery, not being the smartest person in the room.
- **Empowering the team to deliver independently** fosters a stronger, more resilient team and allows individual contributors to grow.
- **A high-functioning team that performs well without direct intervention** is a strong reflection of a manager’s leadership, not a sign of irrelevance.
- **Delegating technical ownership frees the manager to focus on bigger-picture strategic initiatives**, technical direction, and organizational influence.
- **Letting go of ego is essential**—leaders grow by creating impact at scale, not by micromanaging every decision or technical detail.

Coding Round Grades

In this segment, we break down what interviewers really look for in a machine learning coding round. It’s not just about solving the problem—it's about how we think, communicate, and write code that reflects real engineering practices.

- We understand that **weak performance** stems from messy code, poor use of data structures, or inability to explain complexity.
- **Mediocre performance** may reach the right answer but relies heavily on interviewer help, signaling a lack of confidence or strategy.
- A **great performance** involves writing clean, efficient code and clearly explaining trade-offs and scalability.
- To be **exceptional**, we should offer multiple solutions, consider edge cases, and write production-quality code.
- We improve by following a **systematic approach**, thinking out loud, handling edge cases, and prioritizing clean, readable code—just like real ML engineers do in production.

If you want to learn even more from Yayun:

- Book 1 on 1 career guidance with her here: <https://calendly.com/your-friend-yayun/1on1>
- Follow her on LinkedIn: <https://www.linkedin.com/in/yayunjin/>

Hiring Criteria

If you want direct interview support from Alex and Uzair, take their live Maven course with discount code **TARO10**: <https://maven.com/alex-kliotzkin/acing-em-interviews>

This section of the video explains how the engineering manager (EM) interview process differs significantly from that of individual contributor (IC) roles. It emphasizes that EM interviews are more focused on leadership behaviors and decision-making rather than deep technical assessments.

- **Hiring focus shifts significantly:** Unlike IC interviews where 80% of the evaluation is based on technical competence (e.g., DSA and system design), EM interviews flip this ratio — **80% of the evaluation is based on behavioral and leadership skills**.
- **Technical leadership is the focus:** The primary aim is to assess whether a candidate can **lead teams, manage stakeholders, plan roadmaps, and coach effectively**, rather than just solve technical problems.
- **Trust is central:** Behavioral interviews seek to answer one key question: **Can this person be trusted to ensure the team delivers, no matter what?**
- **Technical evaluation still exists but is broader:** The remaining 20% of the EM interview process tests a candidate’s ability to **handle high-level technical topics**, often through system design or architecture discussions rather than deep algorithmic challenges.
- **Successful EM candidates demonstrate strategic thinking and people management**, not just strong coding skills. Knowing how to translate engineering outcomes into business impact is critical.

Talk To Engineers

We explain how to get a real, unbiased understanding of a team by talking directly to engineers—not just the hiring manager.

Here are the core points from the lesson:

- We stress that managers have strong incentives to “sell” their team—they’re evaluated on hiring success and growth—so their perspective is inherently biased and overly positive.

- We recommend explicitly asking to meet 1-2 engineers on the team, ideally ICs who have been there for years and are leaving on good terms, as they provide the most honest, nuanced picture of day-to-day culture, challenges, and growth.

- We emphasize that speaking with the tech lead is especially valuable, since alongside the manager, the TL will influence our experience more than anyone else after joining.

Why Layoffs Happen

We explain why layoffs happen and why they’re a recurring reality in the tech industry.

Here are the core points from the lesson:

- Layoffs occur when companies face bad economics, often triggered by unforeseen macroeconomic shifts like the post-COVID slowdown after years of stimulus-fueled growth.
- Overambitious hiring is a major contributor, especially when overly positive economic conditions lead companies to scale headcount faster than the business can sustainably support.
- Fundamentally unsound business models and poor strategic direction also drive layoffs, and in practice, these factors usually overlap—making layoffs inevitable since executives can’t make perfect decisions all the time.

Taro Book Club: The Pragmatic Programmer - Session 1

If you don't have a Taro Premium membership, you can get it with a discount with Bikram's special link here: [**https://www.jointaro.com/r/bikramk104/**](https://www.jointaro.com/r/bikramk104/)

In this #taro-bookclub discussion, we'll go over Chapter 1 of [**"The Pragmatic Programmer"**](https://www.amazon.com/Pragmatic-Programmer-Journeyman-Master/dp/020161622X) by Andrew Hund & David Thomas

Here's the agenda:

- It's Your Life
- The Cat Ate My Source Code
- Software Entropy
- Stone Soup and Boiled Frogs
- Good Enough Software
- Your Knowledge Portfolio
- communicate

Please send your questions ahead of time to [**Bikram Kashyap**](https://www.linkedin.com/in/kashyapbikram/) in the [**Taro Premium**](https://www.jointaro.com/membership/) Slack, so that they can be included in the discussion. *If you don't have the book or want more details on how this all works, check out the #taro-bookclub channel in the Taro Premium Slack as well!*

Get ready for another engaging and informative session!

General Career Direction Principles

We close with guidance for people who feel completely unsure about where to go next in their tech careers.

Here are the core points from the lesson:

- We suggest that early-career engineers, when possible, start with a big tech company for 1–2 years to gain stability, strong compensation, and a solid learning foundation—then use that platform to explore what they truly want.
- We recommend deliberately experiencing different company archetypes (startup, midsize, big tech) because each environment teaches different skills, exposes us to different cultures, and clarifies our long-term preferences.
- We emphasize that career needs change with age and life complexity—stability tends to matter more in our 30s and 40s—and that most big-tech engineers ultimately transition to smaller companies to find scope and growth.
- We add that when in doubt, choose a company where software engineering is core to the business, not a peripheral function; being in the “engine room” of a product-focused tech company dramatically increases learning, opportunity, and long-term career resilience.

What You’ll Learn

If you want direct interview support from Alex and Uzair, take their live Maven course with discount code **TARO10**: <https://maven.com/alex-kliotzkin/acing-em-interviews>

This course is designed to give you a tactical, insider-level understanding of how to succeed in engineering manager interviews. It focuses on aligning your preparation and responses with what senior leadership is actually looking for in candidates.

- **Understand EM Hiring Criteria:** Learn exactly what directors and senior leaders expect when evaluating EM candidates.
- **Persuasive Interviewing:** Develop strategies to convincingly present yourself to hiring managers at the director level and above.
- **Leadership Storytelling:** Discover which leadership stories to craft and practice to demonstrate your readiness.
- **Answer Optimization:** Learn how to elevate your responses from “good” to “unforgettable” by tailoring them for impact.
- **Practice-Based Learning:** Apply concepts through hands-on exercises to reinforce learning and improve real-time performance.

Paper Reading: Self-supervised Vision Transformers (Foundational Model)

Session led by [Mandar Deshpande](https://www.linkedin.com/in/mandardeshpande1995/)

Dino is a self-supervised learning approach for computer vision tasks. It leverages vision transformers without using any labeled data. Dino v2 introduces several improvements over its predecessor, enhancing performance and scalability. It achieves state-of-the-art results on various benchmarks. Notably, Dino v2 demonstrates strong performance on downstream tasks such as image classification and object detection.

We will be walking through

- Basics of computer vision
- Self supervised learning
- ViT (vision transformers)
- DINO
- DINO v2 model

Paper link - [**https://arxiv.org/abs/2304.07193**](https://arxiv.org/abs/2304.07193)

[Event link](https://www.jointaro.com/event/paper-reading-dino-v2-ie-foundational-model-for-computer-vision/)

Breaking Down Startups

We break down what it really means to join a VC-backed, high-growth startup and how the experience differs from big tech.

Here are the core points from the lesson:

- We outline the major upsides of early-stage startups: massive technical scope, the highest possible career ceiling, and deep fulfillment from directly shaping a company’s destiny—especially when joining at Series B or earlier.
- We explain the risks: early startups are unstable, resource-constrained, and extremely high-pressure, where production issues can threaten the company’s survival and where chaos is normal, not exceptional.
- We show why growth is the single most important signal when evaluating a startup: growth increases technical complexity, headcount, leadership opportunities, and the likelihood of exceptional financial upside; stagnation is a strong red flag.
- We contrast this with agencies/consultancies, noting they rarely offer strong long-term growth for early-career engineers unless someone is already a world-class expert—making startup or big-tech paths much more reliable career foundations.

Technical Bar

If you want direct interview support from Alex and Uzair, take their live Maven course with discount code **TARO10**: <https://maven.com/alex-kliotzkin/acing-em-interviews>

This lesson breaks down how **technical competence** is evaluated during engineering manager (EM) interviews. Although only 20% of the hiring criteria is based on technical skills, understanding how it's assessed is essential for preparation.

- **High-level technical evaluation:** EM interviews typically assess architecture and system-level thinking rather than deep implementation. The goal is to understand if you grasp the *impact and trade-offs* of technical decisions.
- **System design is critical:** Nearly all companies include a **system design round**, as it's a strong signal of your ability to reason about architecture, scalability, and trade-offs at a strategic level.
- **Operational excellence is a focus:** Interviewers may evaluate how you ensure **continuous product integrity** through CI/CD pipelines, quality control, monitoring systems, and delivery processes.
- **Metrics and impact measurement:** Expect questions around **KPIs and OKRs**, especially on how you track team or product success. This reflects your business awareness and leadership maturity.
- **Coding expectations vary by company:** FAANG companies often include **coding rounds** to assess whether you can "get your hands dirty" if needed. Non-FAANG companies may skip this or offer take-home assignments or lightweight coding exercises instead.

Round Structure ML Behavioral

Here’s a summary of what we should expect and aim to convey in the **Project Deep Dive and Behavioral Interview Round**—a conversational but highly valuable part of the ML interview process.

- We’re expected to go beyond surface-level descriptions and **deeply explain a past ML project**, including the problem, our solution, and the technical trade-offs we made.
- This is our chance to show **real-world impact**, how we debugged challenges, optimized performance, or iterated on feedback.
- For **ML infrastructure roles**, we should be ready to discuss tools or systems we built (e.g. training pipelines, feature stores), and how they enabled or scaled team productivity.
- Strong performance in this round depends on **clarity, structure, and storytelling**—connecting technical depth with our thought process and decision-making.
- Ultimately, it’s about showing that we not only *built* things, but thought critically and communicated effectively through the process.

If you want to learn even more from Yayun:

- Book 1 on 1 career guidance with her here: <https://calendly.com/your-friend-yayun/1on1>
- Follow her on LinkedIn: <https://www.linkedin.com/in/yayunjin/>

Common Questions

If you want direct interview support from Alex and Uzair, take their live Maven course with discount code **TARO10**: <https://maven.com/alex-kliotzkin/acing-em-interviews>

In this video, we explore the behavioral side of engineering manager (EM) interviews, which focus heavily on leadership abilities. Each question is designed to assess core leadership competencies, and mastering these areas is critical for success.

- **Hiring & Growing Talent**: Interviewers look for evidence of managing underperformers, nurturing strong performers, and navigating hiring and talent review processes.
- **Delivery Under Pressure**: Candidates must demonstrate the ability to lead high-quality project deliveries on tight timelines, despite resource constraints or setbacks.
- **Stakeholder Management & Influence**: Expect to showcase how difficult feedback was delivered or how stakeholders were influenced for team benefit, often through challenging conversations.
- **Conflict Resolution & Judgment**: Questions will probe how candidates handle internal or external conflicts, as well as how strategic decisions were made with limited data.
- **Vision & Execution**: Interviewers want to hear about setting long-term vision, taking calculated risks, digging deep into complex problems, and operating effectively across all levels.

[Paper Reading] Deploy ML Models from Lab to Production

Session led by: [Akshika Wijesundara PhD](https://www.linkedin.com/in/akshika/)

If you don't have a Taro Premium membership, you can get it with a discount with Akshika's special link here: [**https://www.jointaro.com/r/akshikw802/**](https://www.jointaro.com/r/akshikw802/)

**WHAT YOU'LL LEARN**

- **Discover the benefits of CI/CD pipelines:** Learn how CI/CD pipelines streamline the model building, testing, and deployment processes, enabling rapid iteration and continuous improvement.
- **Containerize your ML code:** Understand the steps and advantages of containerizing your machine learning code to ensure consistency, portability, and efficient resource management.
- **Leverage the AWS stack for ML workloads:** Explore how to effectively utilize AWS as your cloud infrastructure provider to run and scale your machine learning workloads with ease and reliability.

**WHY THIS TOPIC MATTERS**

MLOps isn't just about fancy models - it's about getting them working for real! Learn how to deploy, maintain, and collaborate on ML projects, making you a highly-demand expert who can turn machine learning into real-world impact.

This Facebook Engineering Director Quit To Sail The World

After more than 10 years working at Meta (Facebook), James quit to work on his boat full-time. While sailing, he also contributes to open-source! <https://github.com/tinyplex/tinybase>

Choose A Financially Sound Company

We outline the concrete, proactive steps you can take to reduce layoff risk by choosing the right company upfront.

- The most important controllable factor is company selection, with profitability as a key litmus test, since companies that spend more than they earn are far more vulnerable in tighter markets.
- Hiring quality and pace matter: overly easy interviews and hyper-aggressive headcount growth are red flags that often signal overhiring and future layoffs.
- Healthy companies tend to grow steadily rather than stagnate or attempt extreme, unsustainable expansion, especially given the high cost of tech talent.
- Clear product direction and strong leadership are critical, particularly for startups, and can often be evaluated through interviews, vision clarity, and how leaders communicate.
- There are two valid ways to assess company health—analyzing financials directly or reading secondary signals like employee quality, departures, and trusted insider perspectives—and combining both approaches gives the strongest signal.

Tech Career Growth Videos

Our instructional videos provide software engineers with insights needed for career growth. Masterclasses go incredibly in-depth on a specific tech topic while case studies are real examples from established engineering leaders.