Data Scientist

Emergent builds autonomous coding agents that replace traditional software development by generating, testing,

and deploying production applications directly from plain-language intent. Our systems run in production at global

scale and are used to build millions of real applications.

Since our public launch, we've crossed $100M in ARR and grown to over 10M users across 190+ countries, who

have built 12M+ applications on Emergent. We're backed by Creaegis, Claypond, Sentinel Global, Khosla

Ventures, SoftBank, Google, Lightspeed, Prosus, Together, and Y Combinator.

We're solving the hard part of AI-driven software creation: correctness, reliability, security, and scale in real

production systems. The team is built by repeat founders, Olympiad medalists, IIT & IIM alumni, and leaders from

Google, Amazon, and Dropbox.

We're hiring builders who want ownership, speed, and impact at global scale.

The Role:

Every part of our growth runs on decisions made from data: who activates, who converts, what an agent-built app

actually costs us to produce, and which users are here to build versus here to abuse the free tier. What makes

analytics here unusual is the shape of that data. Alongside the standard event and revenue tables, we sit on an

enormous volume of unstructured signal — millions of agent trajectories (the step-by-step reasoning, actions, and

observations of every build), support tickets, HITL feedback, and the natural-language prompts users write. The

richest insights in the business are buried in that text, and the person in this role is the one who gets them out.

You own the loop end-to-end: what we measure, how we prove it, what the number supports, and what it doesn't.

What You'll Do:

• Turn agent trajectories, support tickets, logs, and user prompts into structured, queryable signal —

summarize-then-embed-then-cluster pipelines (à la Anthropic's Clio / Braintrust Topics): distill each trace

along a dimension with an LLM, embed the summary, cluster and name the patterns, then classify at scale

• Surface early indicators — of confusion, of a coming bug wave, of churn risk, of fraud — that no dashboard

would ever surface on its own, and route them to the right team

• Build predictive models that forecast conversion, retention, expansion, and churn, and embed those

signals directly into product and growth workflows

• Own marketing attribution and MMM: build the media-mix and incrementality models that tell us what's

actually driving signups and paid conversions when per-user attribution is partial and, on mobile, broken by

design

• Own product and growth analytics across the self-serve funnel, web and mobile — activation,

engagement, retention, conversion — and design and analyze A/B and growth tests with real rigor around

power, novelty effects, interference, and causal inference

• Run clustering pipelines over hundreds of thousands of agent trajectories to discover the recurring kinds of

things users try to build and the recurring ways builds fail, then hand product a taxonomy nobody had to

hand-label, along with which clusters predict churn• Model the “aha moment” for new users, including text-derived features from their first prompts and first

agent interactions, and rebuild onboarding around the earliest signals of long-term retention

• Build a gross-margin model that attributes LLM and compute cost down to the individual app and cohort,

and tell product which segments are net-positive

• Untangle a fraud ring that looks anomalous on compute spend but has real payment history, decide

whether it's an enforcement problem or a pricing problem, and defend the call with the data

Who You Are:

• 2 to 5 years in data science or applied ML with a focus on product analytics, growth, or user behavior

• Strong SQL and real comfort working with large, event-level behavioral data at scale

• Solid classical ML foundations — clustering (k-means, HDBSCAN, hierarchical), embeddings and vector

similarity, dimensionality reduction (UMAP/PCA), classification — and a working understanding of when

each is and isn't the right tool

• Genuine skill at deriving insight from unstructured natural-language data — LLM traces, logs, tickets, free

text — and turning it into predictive, queryable signal; familiarity with topic-modeling and trace-clustering

approaches (Clio-style summarize-then-embed pipelines, BERTopic, c-TF-IDF) is a strong plus

• Proficient in Python and the standard data-science stack (pandas, scikit-learn, statsmodels, numpy)

• Data engineering competence — you can design and ship ETL and data models (dbt or equivalent), not

just query what already exists

• Experienced designing and analyzing experiments — sample sizing, power, significance, novelty effects,

interference between tests, and causal methods

• You move fast and go deep, turning around in hours the analysis that takes most people days, because

you've built the intuition to get to the right answer and the discipline to pressure-test it before anyone else

sees it

• You dig past the top-line number to find the confound, you ask whether the metric measures what

everyone assumes it measures, and you never hand over a figure without saying what it supports, what it

doesn't, and what you'd check next

• You use AI agents aggressively to multiply your output, but you treat every AI-assisted result as a draft, not

a deliverable

• Deep curiosity about user behavior and a real instinct for what drives growth, retention, and abuse

• Able to move fluidly between exploratory analysis, ML modeling, hypothesis testing, and crisp strategic

recommendations, and to translate all of it into narratives that drive decisions

Nice to Have:

• Marketing Mix Modeling (MMM), media attribution, or incrementality and geo-testing experience, especially

in low-tracking or post-cookie environments

• Experience at a PLG company with a self-serve funnel and freemium or usage-based / credit-based pricing

• Modern data stack (BigQuery, dbt) and product analytics platforms (PostHog, Amplitude, Mixpanel,

Segment)

• Causal inference methods (difference-in-differences, synthetic control, propensity score matching)

• Fraud, trust and safety, or abuse analytics• Working knowledge of embedding models and vector search, and the practical tradeoffs of running them at

scale

• Familiarity with the economics of AI/LLM products, including COGS modeling where compute is the

dominant variable cost

• You've built or contributed to AI-powered analytical tooling or novel measurement approaches

Ideal For:

• Senior and Staff data scientists who want one of the hardest, most consequential analytics surfaces in AI

software

• Applied ML practitioners who'd rather build the pipeline than wait for one

• Analysts who moved into ML and never stopped shipping

• Anyone drawn to where classical ML, LLM-native insight extraction, margin, attribution, and product all

collide

Benefits and Perks:

1. Daily Meals: Lunch and Dinner provided

2. Family Insurance: 3 Lakhs worth of coverage for you and your family

3. Unlimited Paid Time Off: Take the time you need to recharge and come back refreshed

4. Flexible Working Hours: Work arrangements that fit your life and commitments

Let's build the future of software together.