EchoPath

Inspiration

Students from under-resourced communities often face a double gap: less access to career networks and less visibility into realistic career paths. We built EchoPath to turn “I don’t know anyone in this field” into actionable next steps: find reachable mentors, understand real trajectories, and send a high-quality intro email with confidence.

What it does

EchoPath helps a student discover:

Career paths from people with similar starting points
Mentor matches ranked by multi-factor fit (geography, education tier, hardship context, career goal, level, salary fit)
A generated cold email (kept in English for professional outreach), plus match evidence and icebreaker hooks

It also adds fairness-oriented context:

Hardship score explanation panel (top contributing factors + recommendation impact)
Reachable vs. Aspirational recommendation modes
Local support suggestions and prototype fairness checks

How we built it

We built a full-stack web app:

Frontend: Next.js (React + TypeScript), Tailwind CSS, Framer Motion, React Flow
Backend: FastAPI + Pydantic
Data/Scoring: Python (Pandas/NumPy/scikit-learn style weighted scoring), ZIP/FIPS mapping, hardship scoring
LLM pipeline: Rapidfire/Gemini/OpenAI fallback chain for email generation and path-related language output
Storage/RAG (prototype-ready): local JSON/CSV and optional PostgreSQL + pgvector setup

We implemented practical product details for demo reliability:

U.S. ZIP validation and handling
Diverse path generation to reduce repetitive outputs
Mentor thresholding and cap for quality (e.g., minimum fit cutoff)
English/Spanish UI language switch (UI localized; cold email remains English)

Challenges we ran into

Data realism vs. demo stability: balancing prototype data with believable outputs
Path diversity: early outputs were repetitive, so we added diversity logic and anti-repeat ranking
Fairness UX clarity: turning a single hardship number into an explainable, actionable panel
Localization scope: ensuring the UI is multilingual while keeping outreach email language professional (English)

Accomplishments that we're proud of

Built an end-to-end fairness-aware career navigation experience
Added interpretable matching signals instead of a black-box score only
Improved output quality with path diversity and mentor-fit constraints
Delivered bilingual UI support with minimal friction for users

What we learned

Explainability dramatically improves trust in recommendation systems
Product framing matters as much as model quality in social-impact tools
Small UX decisions (copy, defaults, interaction stability) strongly affect usability
Fairness features should include both diagnosis and intervention guidance