Describe a failure and a success
Company: Reddit
Role: Machine Learning Engineer
Category: Behavioral & Leadership
Difficulty: medium
Interview Round: Onsite
## Questions
1. Tell me about a time you **failed** (or something didn’t go as planned). What happened, what did you learn, and what would you do differently?
2. Tell me about a time you were **successful**. What was your specific contribution and what was the impact?
## Expectations
- Use concrete examples from engineering/ML work (delivery, incident, model launch, collaboration).
- Highlight ownership, communication, and measurable outcomes.
- Include reflection: what you changed afterward.
Quick Answer: This question evaluates ownership, communication, accountability, and impact assessment within engineering and machine learning contexts, focusing on incident management, model launches, delivery outcomes, and collaboration.
Solution
### Use a clear structure: STAR / CAR
A reliable format is **STAR**:
- **S**ituation: context and stakes
- **T**ask: your responsibility (not the team’s)
- **A**ction: what you did, decisions you made, tradeoffs
- **R**esult: measurable outcome + what you learned
Or **CAR** (Context–Action–Result) if you want to be more concise.
### 1) Failure story: what interviewers look for
They want:
- A real failure (not a disguised humblebrag)
- Ownership without self-blame spirals
- Specific corrective actions and lasting process change
#### Good failure examples (pick one)
- Shipped a model without proper offline/online parity checks → online metrics regressed.
- Underestimated data quality issues → training pipeline produced silent corruption.
- Poor stakeholder alignment → built the wrong thing / mis-scoped MVP.
#### What to include
- Your decision point: why you chose that path with the info you had.
- Detection and response: how you triaged, communicated, and mitigated.
- Prevention: what you changed (tests, monitoring, review checklist, rollout plan).
**Concrete additions that make it strong**:
- Add numbers: “CTR dropped 1.2% relative”, “p95 latency +40ms”, “2-hour incident”.
- Add a process fix: canary release, automated data validation, feature store adoption, postmortem template.
### 2) Success story: what interviewers look for
They want:
- Clear individual contribution and leadership (even without formal authority)
- Impact: business + technical metrics
- Sound decision-making: tradeoffs, prioritization, and execution
#### Good success examples
- Launched a ranking model or retrieval improvement with measurable lift.
- Reduced training cost/latency significantly.
- Built an experimentation framework or logging pipeline that unblocked multiple teams.
#### What to include
- Scope: why it mattered.
- Your role: design decisions, driving alignment, unblocking others.
- Outcome: A/B results, reliability improvements, developer productivity.
### 3) Handling follow-ups
Common follow-ups and how to answer:
- “What would you do differently?” → name 1–2 specific changes you now apply.
- “What did you learn?” → tie to a principle (e.g., ‘validate assumptions early’, ‘instrument before optimizing’).
- “How did you influence others?” → show communication cadence, docs, design reviews.
### 4) Quick prep checklist (so you don’t ramble)
Prepare 2 stories (failure + success), each with:
- One-sentence summary
- 3 actions you personally took
- 2 metrics of impact
- 1 lasting change you implemented
This consistently produces senior-level behavioral answers.