PathForge

Career transition advice is abundant but not actionable. Turn generic guidance into a persistent, adaptive roadmap that evolves with you.

Comment

Inspiration

Career transitions are often guided by generic advice that does not match a person's background, time constraints, or learning style. PathForge was inspired by the idea that people should get a roadmap that feels personal, actionable, and adaptive instead of one-size-fits-all.

What it does

PathForge helps users plan and execute a career transition through:

  • Profile intake (text, LinkedIn URL, optional resume upload)
  • AI-driven diagnostic interview questions
  • Personalized roadmap generation with weekly phases and tasks
  • Real-world local event recommendations integrated into the roadmap
  • Task-level check-ins with reflections, blockers, and time tracking
  • Voice reflection transcription (speech-to-text) for faster progress logging
  • Adaptive week optimization based on completed work and feedback
  • Reflection-based roadmap updates and progress history views

How we built it

  • Frontend: React + TypeScript + Vite + Tailwind + Framer Motion
  • Backend: Node.js + Express + TypeScript
  • AI stack:
    • Google Gemini (gemini-2.5-flash) for interview, roadmap, adaptation, and analogy generation
    • Gemini search tool for finding relevant local events
    • External STT batch API for voice reflection transcription
    • Braintrust logging for observability
  • Data handling:
    • Client-side persistence with localStorage for profile, interview context, roadmap, and check-ins
    • SQLite schema scaffolded for future server-side persistence

Challenges we ran into

  • Getting reliable structured JSON output from LLM responses and handling malformed/partial responses safely
  • Managing adaptation logic while preserving task IDs and schema consistency
  • Integrating voice recording/transcription flow smoothly into the task check-in UX
  • Keeping frontend and backend contracts stable while iterating quickly on roadmap/task schemas
  • Avoiding empty or low-quality generations by designing robust fallback behavior

Accomplishments that we're proud of

  • Built an end-to-end product flow from onboarding to adaptive execution
  • Implemented meaningful roadmap adaptation driven by actual user check-ins
  • Added practical task support features (analogies, checklists, resource links, progress history)
  • Combined AI generation with real event discovery to ground plans in the real world
  • Shipped a polished, modern UX with clear progression and feedback loops

What we learned

  • Prompt design and output constraints are critical for predictable product behavior
  • Schema-first thinking reduces breakage when multiple AI features evolve in parallel
  • Local-first persistence accelerates prototyping but highlights the need for durable backend state later
  • User trust improves when AI outputs are contextual, specific, and revisable based on feedback
  • Observability for AI workflows is essential for debugging and iteration speed

What's next for PathForge

  • Move persistence from localStorage to fully integrated backend DB models and user accounts
  • Add authentication and multi-device sync
  • Improve interview voice flow with true speech capture/transcription in all interview steps
  • Add stronger validation/guardrails for AI outputs and richer fallback strategies
  • Introduce role-specific roadmap templates and benchmark milestones
  • Expand analytics and coaching insights (consistency trends, risk flags, estimated readiness)
  • Prepare deployment-ready env management and production API configuration

Built With

Share this project:

Updates