Pathforge

Career exploration is a fragmented loop today: one tool drafts the resume, another suggests courses, a third tracks applications, a fourth simulates interviews. None of them know about each other. The user is the integration layer.

An exploratory build of a single surface that takes a goal (e.g. 'land a senior PM role at an AI infra company') and returns a personalised plan: skills to close, projects to ship, applications to send, conversations to have — all updated as the user makes progress.

Multi-agent orchestration layer running on top of frontier LLMs. Each agent owns a slice of the loop (skill gap analysis, learning path, application coach, interview prep). Memory persists across sessions in a vector store so the system learns the user.

• Next.js 15 App Router · TypeScript · Tailwind for the surface.
• Supabase + pgvector for memory and document storage.
• Anthropic + OpenAI behind a router that picks the model per task.
• Background workers for nightly job-board ingestion and personalised digest emails.
• Per-user privacy: user data never leaves the row-level-secured Postgres, no third-party analytics.

Personal exploration. Built primarily for myself, shared as a writeup. Not a commercial product.

The hard part is not the LLM. It is making the system feel like one continuous coach instead of seven good chat windows.