Edupath

main dashboard
recommendation
sub-dashboard (planning)
boarding
recommendation-architecture
overall-architecture
plan-architecture

Inspiration

College planning shouldn't depend on privilege. While wealthy families pay thousands for private counseling, first-generation and under-resourced students often navigate complex milestones alone. We built EduPath to democratize high-tier academic guidance, converting elite counseling into a scalable, open-access serverless advisor.

What it does

EduPath turns a student's unique background into concrete, actionable steps:

Onboarding: Ingests student parameters (grade, interests, goals) into a persistent profile.
AI Recommendations: Generates three tailored development tracks with custom difficulty scores.
Action Plans: Builds comprehensive step-by-step milestone checklists and an 11-12th grade College Fit Chart.
Dynamic Memory: Tracks real-world success or failure outcomes to automatically recalibrate future recommendations.

How we built it

Frontend: React 19, TypeScript, Vite, and Tailwind CSS hosted globally on AWS Amplify.
API Entrypoint: Amazon API Gateway proxying REST endpoints directly to backend logic.
Compute Layer: Decoupled AWS Lambda microservices running Python 3.12.
AI Brain: Amazon Bedrock invoking Claude via the structured Messages API, configured with low-temperature constraints for pure JSON parsing.
Data & Storage: Amazon DynamoDB leveraging composite keys (userId/trackId) for rapid session updates, alongside Amazon S3 for long-form checklist retention.

Challenges we ran into

Lambda Path Sandbox: Solved local-to-cloud prompt template pathing crashes (FileNotFoundError) by implementing dynamic directory resolution using runtime environment parameters.
Nested Database Mutations: Prevented multi-row data collision when writing to sub-arrays by implementing atomic DynamoDB UpdateExpression logic with list_append.
API Layout Migration: Re-engineered backend loops from legacy string completion patterns into structured target array formats required by the modern Bedrock syntax.

Accomplishments that we're proud of

Unified two independent backend blocks (Recommendation and Planning) seamlessly over a single database pattern.
Achieved a 100% success rate across our 41-hook local verification suite before shipping our code to production.
Completed full end-to-end integration from our Amplify web interface down to real live data writes.

What we learned

Context Sharing: Packaging code dependencies inside specialized serverless deployment layers to optimize cold-start performance.
Deterministic AI Generation: Structuring exact prompt engineering payloads to completely restrict LLM responses to strict, machine-readable JSON arrays without prose.

What's next for Edupath

Amazon Textract: Processing unofficial high school transcript uploads to automatically extract GPAs, credits, and course history.
AWS Step Functions: Transitioning API generation sequences into robust serverless state-machine workflows.
Amazon SES: Deploying an automated email notification system to dispatch checklist alerts and financial aid deadlines directly to parents.