PathFinder AI

An AI career advisor that interviews you across 12 dimensions and delivers scored, actionable career recommendations in real time.

Comment

PathFinder AI

The Big Problem

Career decisions are among the highest-stakes choices people make, yet most people navigate them with Google searches, generic quizzes, or expensive career coaches. Existing tools either reduce you to a personality type or overwhelm you with unfiltered job listings. Nobody connects who you are to what you should do in a structured, evidence-based way.

What We Built!

PathFinder AI is a full-stack career advising application with three stages:

1. Conversational Intake (12 dimensions)

A chat-based interview that progressively builds a structured career profile across:

  • Interests, values, and working style
  • Hard and soft skills
  • Risk tolerance and financial constraints
  • Geographic flexibility and education
  • Timeline urgency
  • Purpose/meaning priorities and burnout concerns

Each answer is parsed in real time into structured data — the profile builds visibly as you talk.

2. Multi-Agent Analysis

The completed profile is sent to a multi-agent backend (Fetch.ai uAgents) where specialized agents evaluate career fit across different angles: skill matching, values alignment, market demand, salary viability, and burnout risk. Progress streams back via SSE.

3. Scored Recommendations

Results are presented as career cards with:

  • Fit score (0–100%) showing how well the role matches
  • Why it fits — specific reasons tied to your profile
  • Watch out for — honest concerns
  • Next steps — concrete actions: certifications, companies to target, portfolio projects
  • Salary range — market-based compensation data

Architecture

React SPA  ←→  Fastify API Gateway  ←→  Python Agent Service
   (Vite)        (TypeScript)           (Fetch.ai uAgents)
                     ↕
                SQLite (Drizzle ORM)
  • Frontend: React 19 + TypeScript + TanStack Query + React Router + Zod validation
  • API: Fastify 5 + TypeScript + Drizzle ORM + SQLite + SSE streaming
  • Agent service: Python + Fetch.ai uAgents framework (multi-agent orchestration)
  • Contract: Zod schemas shared between frontend and API — every request and response is validated at both ends

Technical Highlights

  • Deterministic fallback engine: The intake works without any LLM — a scripted question sequence with regex-based extraction ensures the demo always works, even if the agent service is down
  • State machine with guards: Sessions follow strict transitions (intake → analyzing → complete) with validation at every step. Invalid transitions return clear errors.
  • Self-test endpoint (/ready): Runs a full end-to-end golden path internally and reports a structured checklist — hit it once before a demo to guarantee everything works
  • 29 integration tests covering the complete happy path and all error paths
  • Observability: Every response carries X-Request-Id and X-Response-Time headers; errors include request IDs for tracing
  • Demo mode: DEMO_MODE=true enables operator tools, backup scenarios, and deterministic SSE timing without needing the Python agent service

What Makes This Different

  1. Structured, not vibes: 12 explicit dimensions, not a vague "what's your personality type"
  2. Multi-agent, not single-prompt: Different agents specialize in different aspects of career fit
  3. Actionable output: Not just "you'd be good at X" — specific companies, certifications, portfolio projects, and salary ranges
  4. Honest concerns: Every recommendation includes what might go wrong, not just why it's great
  5. Real-time streaming: Watch the analysis happen live, not a loading screen then a dump

Team

Jonathan Ty, Sammy Hamouda DiamondHacks 2026 — Built in 24 hours with love, sweat, tears, and an inhumane amount of Celsius©

Built With

Share this project:

Updates