Alina Chen

Painpoint: Engineers increasingly work alongside AI tools such as Claude code, the reasoning trace behind every change (the constraints considered, the dead ends hit, the trade-offs chosen) disappear between sessions and handoffs. Agents become black boxes, teams rediscover the same constraints, and nobody can compare how two engineers (or two agents) approached the same problem.

Solution: Inspired by GitHub, built for collaboration between engineers and their AI coding tools, with a shared reasoning layer capturing constraints, dead ends, and trade-offs across human + agent workflows.

What it does:

• A CLI wraps git with commands like ckpt add / commit / push that capture reasoning alongside each change in local SQLite and sync to the API on push
• An MCP server lets Claude Code read and write that reasoning natively during a coding session, so capture happens in the flow
• Web app UI: a dashboard of recent checkpoints across repos, step timelines with code diffs at each decision, a side-by-side compare view for two engineers or agents solving the same problem, and threaded comments for team review
• Teams and repos are scoped through GitHub OAuth, with an open webhook protocol so any AI coding tool can push records in

Painpoint: Increasing FOMO-driven, gamble-like behavior among new investors, and the fatigue from the gap between textbook finance and real markets. All leads to avoidable losses and emotional frustration around personal finance.

Solution: A full-stack investment education app where users browse curated investment themes, research companies, and learn through an AI advisor that tracks progress. The goal is to build conviction before users invest.

What it does:

• Semantic search across themes and filings via a Pinecone-powered RAG layer
• Live catalyst feed tied to company profiles
• Scoring engine for personalized learning paths, targeting knowledge gaps

End-to-end ownership: auth, onboarding, streaming LLM chat, RAG, data visualization, and persistent progress, live in production with users.

Built-in eval loop: after each chat, a background function grades the conversation against a rubric and logs dimension-level scores to the database.