Highlight Background and Impactful Projects in Self-Introduction

# How to Craft a Strong 60–90s Self-Intro (DS, Technical Screen) ## Structure (Simple Formula) - Present → Past → Proof → Fit - Present: Who you are now and focus areas. - Past: Relevant experiences/skills at scale. - Proof: 1–2 mini STAR stories with quantified results. - Fit: Why this role/team now. ## Mini STAR in One Sentence - Situation/Task: What problem or goal. - Action: What you specifically did (methods, systems, collaboration). - Result: Measurable outcome with numbers and guardrails. ## Sample 75–90 Second Answer (Tailored to consumer-scale DS) "I’m a data scientist with 4 years of experience in product experimentation and recommendation systems, focusing on ranking, causal inference, and shipping models to production at scale. Most recently at a consumer app with >50M DAU, I owned experiment design and model improvements for the home feed. Two examples: First, we had a cold-start relevance gap for new users. I partnered with infra and built a two-tower retrieval model with user/content embeddings and approximate nearest neighbors. We reduced p50 retrieval latency by 35 ms and lifted day-1 watch time by 4.8% in an A/B test across 5% traffic, with no increase in complaint rate. Second, creator churn spiked after policy changes. I built uplift models and a causal segmentation analysis to target high-risk cohorts, then ran a staged experiment on tailored notifications. We reduced 4-week churn by 7.2% for the targeted segment and improved Gini uplift by 0.11. I’m excited about tackling large-scale ranking and experimentation problems, working end-to-end from data to deployment, and collaborating with engineers and PMs to move metrics that matter." ## Fill‑In Template (Customize Quickly) - Present: "I’m a [title] with [X] years in [domains: experimentation, recsys, NLP, trust & safety], focused on [methods: causal inference, embeddings, uplift, bandits] and shipping impact at scale." - STAR 1: "We faced [problem/metric goal]. I [action: model/method, system, cross‑team collab]. Result: [metric + magnitude + guardrail]." - STAR 2: "Additionally, [problem]. I [action]. Result: [metric]." - Fit: "I’m excited about [team’s problem space], bringing [skills] to drive [target metrics] while collaborating cross‑functionally." ## Quantification Tips - Prefer business or user metrics: retention, session length, watch time, creator churn, safety rates, revenue, latency. - State sample sizes/traffic and guardrails when possible: "+3.1% retention at 20% traffic; no regressions in latency or reports." - Express both relative and absolute when meaningful: - "+4.8% watch time" or "+0.9 min/session" - "-35 ms p50 retrieval latency" ## Common Pitfalls - Too long or vague; aim for 130–200 words (~60–90s at normal pace). - Listing responsibilities instead of outcomes. - Tech name-drops without why/how/impact. - No connection to the role/team. ## Quick Validation Checklist - Time your delivery to under 90 seconds. - Each project line has: problem → your action → number. - Replace internal code names with generic descriptions; keep confidentiality. - Have a 60s version (1 project) and a 90s version (2 projects) ready. ## 60-Second Variant (One Project) "I’m a data scientist with 4 years in experimentation and ranking for consumer feeds. Recently, I led a cold‑start relevance effort: built a two‑tower retrieval model with ANN search, partnering with infra to keep p50 latency under 100 ms. In a 5% A/B, we lifted day‑1 watch time by 4.8% without raising complaint rate. I enjoy shipping pragmatic ML with strong experiment design and clear guardrails, and I’m excited to apply that to large‑scale ranking and measurement problems on this team." Use this structure, swap in your own metrics and techniques, and rehearse until it’s crisp and natural.