Data Scientist (Python / LLM / MLOps) - Full Time - US Only

The Mission

Turn messy public-record & web data into trustworthy mass-tort signals - via rapid EDA, rigorous modeling, and production-grade APIs - so lawyers act on facts, not hunches.

What You’ll Own

• EDA Autopilot - Grab raw JSON/CSV/HTML, profile it, spot outliers, and surface “aha” patterns - without waiting for a PM to ask.
• Model Builder - Train and tune classification / ranking models (Typical classifiers, light ML, LLM-based RAG) that lift recall & precision week-over-week.
• API Integrator - Package models behind FastAPI endpoints, validate / marshal schemas with Pydantic, and push to GitHub Actions CI in a day.
• MLOps Wrangler - Monitor drift, schedule batch recall and write lightweight tests.
• Insight Storyteller - Ship clear notebooks / dashboards & concise Loom walk-throughs that legal SMEs grok in minutes.
• Startup Swiss-Army Knife - Spot gaps (data gaps, labeling gaps, infra gaps) and patch them before anyone asks. Ambiguity is the default.

What Success Looks Like

• Week 4 - First exploratory notebook flags a recall-worthy defect the founders didn’t see.
• Week 6 - A FastAPI route serving the trained model hits < 300 ms P95 latency in prod.
• Quarter 1 - Recall ↑ 15 pp and false-positive rate ↓ 10 pp on live web-scrape feed; zero pager-alerts.

Your Toolkit

• 2-4 yrs Python data science: pandas/Polars, Astral stack, PyTorch/TF.
• Comfort spinning up FastAPI + Pydantic micro-services.
• Familiarity with using rate-limited LLMs to augment and clean existing datasets
• Solid SQL & object-storage chops (Postgres, DuckDB, S3).
• CI/CD familiarity (GitHub Actions or similar); basic IaC a plus.
• You document & demo your work proactively - no babysitting required.

Nice-to-Haves

• Prior scraping work (Scrapy, Playwright) or PACER/NHTSA/FDA datasets.
• Experience with vector DBs (Qdrant, pgvector) & prompt-engineering.
• Exposure to SOC 2 or other regulated-data environments.

Interview Process

• 15 minute initial overview call
• ~2-3 hour take home assessment with dataset provided. EDA and written communication expected
• 1 hour pair programming assessment with CTO, with screenshare and agent use
• 30 minute Q&A with founding team

Why Join

Green-field ML canvas, instant customer feedback, and exposure to leadership in a venture-backed startup already generating revenue and real impact in legal tech. Shape how mass-tort intelligence is built in the AI era.

Ready to build? Apply today.