TalentOS Story

Video demo: https://cln.sh/F3jHrZKg

Inspiration

Recruiters waste 40+ hours manually piecing together LinkedIn profiles. Asking "Find someone who worked at OpenAI AND Anthropic" shouldn't require spreadsheet archaeology. We wanted to make human capital queryable like a database.

What it does

TalentOS lets you search LinkedIn semantically. Query "dual-unicorn alumni" or paste 5 URLs for a promotion decision—get instant matches and chat with deep career context. Ask "Who's a scaling risk?" and get behavioral answers, not keywords.

How we built it

• Lightpanda scrapes LinkedIn profiles 10x faster (no CSS rendering overhead)
• Claude agents extract 100+ behavioral facts per profile through

iterative prompting

• Sanity Transform structures raw data into semantic schemas with career patterns, risk flags, and experience graphs
• RAG chat powered by vectorized facts stored in Sanity's Content Lake

Challenges we ran into

Balancing speed vs depth—scraping 50 profiles in 5 minutes while running multi-pass LLM enrichment without exploding API costs. Solved with tiered processing and Sanity's efficient storage.

Accomplishments that we're proud of

Complex queries like "0→1 founders who exit at PMF" work instantly. Chat understands context: "This person's burnout risk?" → detailed behavioral analysis.

What we learned

Structured semantic storage beats vector-only RAG. Sanity's schema + GROQ queries let us combine exact filters with semantic search—impossible with pure embeddings.

What's next for TalentOS

Team synergy scoring, automated reference question generation, and predictive retention modeling based on career pattern analysis.

Built With

• claude
• lightpanda
• sanity
• typescript