← Meta Interview Insights

Meta·Machine Learning Engineer·Technical Phone Screen·Senior

Senior
May 2026

Summary

Technical screen at Meta for an ML Engineer role, focused almost entirely on LLM work. Two meaty topics: agent evaluation and post-training pipelines. The questions were specific enough that vague answers probably wouldn't fly.

Questions Asked (5)

Q1

Walk me through a recent project you worked on involving LLM agent evaluation. What tasks or benchmarks did you design, and what metrics did you use?

A/B Testing & ExperimentationTechnical Trade-offsProduct Analytics & Metrics
Author's notes

This one took up a lot of airtime.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Select a concrete LLM agent project you led, and structure your answer around the problem, the evaluation design, and the outcomes. Emphasize how you chose tasks and metrics to balance coverage, cost, and alignment with product goals, and quantify the impact of your evaluation.

Pro tip: Show that you think about evaluation as a product: discuss how you prioritized metrics based on business impact and iterated on benchmarks as the agent evolved. Mention any trade-offs you made between automated metrics and human evaluation.

1. Set the context

Briefly describe the project, the agent's purpose, and why evaluation was critical. Mention the stage (e.g., pre-launch, iteration) and your role.

2. Design tasks and benchmarks

Explain how you selected or created tasks that represent real user scenarios. Cover diversity, difficulty, and any synthetic data generation or curation methods.

3. Choose metrics

List the metrics used (e.g., task success rate, tool call accuracy, latency, cost) and justify why they matter. Include both automated and human evaluation where relevant.

4. Run experiments and analyze

Describe how you executed the evaluation, including any A/B tests or comparisons. Highlight key findings and how they informed decisions.

5. Share outcomes and learnings

Summarize the impact (e.g., improved success rate, reduced cost) and what you would do differently. Connect to broader lessons about agent evaluation.

Key Points to Mention

  • Task design: coverage of edge cases, multi-turn interactions, and tool use
  • Metrics: task success rate, tool call accuracy, latency, cost per task, and human preference scores
  • Automated vs. human evaluation: trade-offs and when to use each
  • A/B testing or online evaluation to validate offline benchmarks
  • Iterative benchmark refinement based on agent failures and user feedback
  • Quantified impact: e.g., 'reduced task failure rate by 20%' or 'cut evaluation cost by 30%'

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q2

How did you address stochasticity in agent evaluation, and what did you do about judge bias when using an LLM as a judge?

A/B Testing & ExperimentationTechnical Trade-offs
Author's notes

Honestly the judge bias part is where I fumbled a bit.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by acknowledging that stochasticity and judge bias are inherent challenges in agent evaluation, then describe a systematic approach to quantify and mitigate both. Emphasize the use of statistical methods (e.g., multiple runs, confidence intervals) and bias mitigation techniques (e.g., calibration, human validation) to ensure reliable evaluation. Conclude with lessons learned and how you balanced trade-offs between rigor and efficiency.

Pro tip: Quantify the impact of stochasticity and bias on your metrics—showing you measure and manage these issues demonstrates maturity beyond just acknowledging them.

1. Acknowledge and Define the Challenges

Clearly state that stochasticity arises from random seeds, environment variability, and agent policy randomness, while judge bias stems from LLM's inherent biases (e.g., position, verbosity, self-enhancement).

2. Design Robust Evaluation Protocol

Describe running multiple trials with different seeds, using confidence intervals, and ensuring sufficient sample size to distinguish signal from noise. For judge bias, outline steps like randomizing order, using multiple judges, and calibrating against human labels.

3. Implement Bias Mitigation Techniques

Detail specific methods: for stochasticity, use variance reduction techniques (e.g., common random numbers) and report distributions; for judge bias, employ debiasing prompts, ensemble judges, and regular human audits.

4. Validate and Iterate

Explain how you validated the evaluation pipeline (e.g., correlation with human judgments, A/B tests on judge variants) and iterated to improve reliability.

5. Summarize Impact and Learnings

Conclude with the outcomes: improved evaluation reliability, reduced bias, and key takeaways for future work, highlighting trade-offs made.

Key Points to Mention

  • Multiple runs with different random seeds to estimate variance and confidence intervals.
  • Use of statistical tests (e.g., bootstrap, t-test) to determine significance of differences.
  • Judge bias types: position bias, verbosity bias, self-enhancement bias, and how to detect them.
  • Mitigation strategies: randomizing order, using multiple LLM judges, calibration with human annotations.
  • Human-in-the-loop validation to periodically check LLM judge alignment.
  • Trade-offs between evaluation cost, latency, and reliability.

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q3

Describe how you prepared or synthesized training data for an SFT or DPO pipeline. What were your sources, filtering criteria, and how did you handle deduplication and preference-pair construction?

Data ModelingTechnical Trade-offsSystem Design
Author's notes

Big open-ended question.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Structure your answer around a specific project, detailing the end-to-end data pipeline from source collection to final dataset. Highlight key decisions, trade-offs, and metrics that demonstrate rigor and impact. Emphasize how your approach aligns with Meta's scale and quality standards.

Pro tip: Quantify the impact of your data preparation on model performance (e.g., 'improved win rate by X%') and mention any automated tools or frameworks you built to streamline the process, showing scalability.

1. Context and Objective

Briefly describe the project, the model's goal, and why high-quality training data was critical. Mention the scale (e.g., number of examples) and the specific SFT or DPO pipeline.

2. Data Sources and Collection

List the sources (e.g., human annotations, existing logs, synthetic data) and how you ensured diversity and coverage. Explain any sourcing challenges and how you addressed them.

3. Filtering and Quality Control

Detail your filtering criteria (e.g., heuristics, model-based filters) and how you balanced quality vs. quantity. Mention any iterative refinement based on model feedback.

4. Deduplication and Preference Pair Construction

Explain your deduplication methods (e.g., MinHash, embeddings) and how you constructed preference pairs for DPO, including strategies for hard negatives and ensuring label quality.

5. Evaluation and Iteration

Describe how you evaluated the dataset's impact on model performance and iterated on the pipeline. Highlight any metrics (e.g., win rate, accuracy) and lessons learned.

Key Points to Mention

  • Use of both human and synthetic data, with clear criteria for each
  • Filtering techniques such as perplexity thresholds, toxicity filters, and length constraints
  • Deduplication methods like MinHash, SimHash, or embedding-based clustering
  • Preference pair construction for DPO: ensuring chosen/rejected pairs are meaningful and diverse
  • Handling of class imbalance and bias in the data
  • Automation and scalability of the pipeline (e.g., using Spark, Ray, or custom tools)

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q4

What were the hardest challenges you faced in post-training, specifically around reward hacking, length bias, or evaluation flakiness?

Root Cause AnalysisTechnical Trade-offsA/B Testing & Experimentation
Author's notes

Length bias came up and I had a war story about it which landed well.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Pick one specific challenge (e.g., reward hacking) and walk through it using a structured problem-solving narrative: context, symptoms, root cause analysis, experiments, solution, and impact. Emphasize the technical trade-offs and how you validated the fix with rigorous evaluation.

Pro tip: Quantify the impact of your solution (e.g., reduced reward hacking incidents by X%, improved evaluation stability by Y%) and mention how you balanced short-term fixes with long-term robustness. Also, show awareness of Meta's scale and the need for automated detection.

1. Set the Context

Briefly describe the post-training setup (e.g., RLHF, reward model, evaluation pipeline) and why the challenge mattered for the product or research goal.

2. Identify the Challenge

Clearly state the specific issue (reward hacking, length bias, or evaluation flakiness) and provide concrete symptoms (e.g., model outputs gaming the reward, inconsistent eval scores).

3. Root Cause Analysis

Explain how you diagnosed the root cause using data analysis, ablation studies, or debugging tools, and what you found (e.g., reward model overfitting to length, noisy evaluation data).

4. Solution and Trade-offs

Describe the solution you implemented (e.g., reward shaping, length penalty, evaluation protocol changes) and discuss trade-offs (e.g., impact on model performance, computational cost).

5. Results and Learnings

Share the outcome with metrics (e.g., reduced hacking, stabilized evals) and key takeaways that demonstrate growth and technical depth.

Key Points to Mention

  • Reward hacking: reward model over-optimization, specification gaming, and mitigation techniques like reward shaping or adversarial training.
  • Length bias: how models exploit length as a proxy for quality, and solutions like length normalization or penalization.
  • Evaluation flakiness: sources of variance (e.g., sampling, prompt sensitivity) and methods to reduce it (e.g., multiple runs, confidence intervals, robust metrics).
  • A/B testing and experimentation: how you designed experiments to validate fixes and measure impact.
  • Technical trade-offs: balancing model performance, computational efficiency, and robustness.
  • Meta-specific context: scale, cross-functional collaboration, and production impact.

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q5

How did you measure success in your post-training work, and what did a meaningful improvement actually look like?

Product Analytics & MetricsA/B Testing & Experimentation
Author's notes

Short answer: be ready to talk about both automated metrics and human eval, because they'll ask about both.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by defining the specific post-training objective (e.g., improving model quality, reducing latency, or increasing user engagement) and the metrics you chose to track it. Then walk through how you established baselines, designed experiments or evaluations, and interpreted results to determine if the improvement was meaningful. Emphasize the link between offline metrics and online business impact, and how you iterated based on findings.

Pro tip: Meta values a strong hypothesis-driven approach: always tie your success metric to a clear product or business goal, and be ready to explain why you chose that metric over alternatives. Also, mention how you guarded against metric gaming or false positives (e.g., via guardrail metrics and statistical significance).

1. Define the objective and success metric

Clearly state the post-training goal (e.g., improve model accuracy, reduce inference cost, increase user engagement) and the primary metric you used to measure it. Explain why this metric aligns with business or product goals.

2. Establish a baseline and set targets

Describe how you measured the current performance (baseline) and what target you set for meaningful improvement. Mention any domain knowledge or historical data that informed the target.

3. Design the measurement approach

Explain whether you used offline evaluation (e.g., holdout sets, cross-validation) or online A/B testing, and how you ensured statistical rigor (e.g., power analysis, significance testing).

4. Analyze results and validate impact

Share the results: how much did the metric improve? Was it statistically significant? Did you check for guardrail metrics or unintended consequences? Describe how you validated that the improvement was real and not due to noise.

5. Iterate and learn

Discuss what you did next: did you ship the change, iterate further, or abandon it? Highlight any lessons learned about the metric or the model that informed future work.

Key Points to Mention

  • Choice of primary metric (e.g., accuracy, F1, latency, CTR) and its connection to business impact
  • Use of offline evaluation vs. online A/B testing, and why one was chosen
  • Statistical significance, confidence intervals, and sample size considerations
  • Guardrail metrics to ensure no regression in other areas
  • Quantifiable improvement (e.g., 'increased F1 by 5% relative' or 'reduced latency by 20ms')
  • Iterative process: how results informed next steps or model iterations

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.