This is a submission for the MLH x DEV Writing Challenge
What I Built
Researcher Campus — a full-stack, AI-powered academic research pipeline that replaces 5+ disconnected research tools (novelty checkers, citation managers, compliance checklists, drafting tools) with a single 7-stage workflow: novelty check → gap analysis → manuscript drafting → compliance audit.
The core problem: researchers waste hours jumping between tools to check if their idea is novel, find literature gaps, draft a proposal, and then manually verify it meets submission standards. Researcher Campus automates that entire chain.
Key pieces:
- A real-time novelty-detection engine that queries 5 academic APIs simultaneously (Crossref, arXiv, OpenAlex, Semantic Scholar, Europe PMC) and compares results using 768-dimension vector embeddings, with a TF-IDF fallback for offline/degraded scenarios
- A 5-tier Gemini fallback chain with auto-failover, so proposal generation doesn't break if one model tier is rate-limited or down
- A 4-guard compliance auditor checking citations, anonymity, formatting, and tone before submission
Demo
🔗 Live: researcher-campus.vercel.app
💻 Code: github.com/jayesh-thar/Researcher-Campus
Tech Stack
- Backend: 31 REST endpoints, JWT auth, AES-256-GCM encryption, ~7,000 req/sec sustained with 0 errors under load
- Infra: Docker on AWS ECS Fargate, provisioned via Terraform, GitHub Actions CI/CD
- Data: MongoDB + an RDS-based audit-logging layer
- Quality: 21 automated tests, 0 TypeScript errors
Hackathon Experience
I didn't build this at a live hackathon — Researcher Campus started as a personal problem I kept running into: jumping between five different tools just to check if a research idea was novel. After digging into how existing tools approached this, I decided to build the whole pipeline myself, solo, working online at my own pace.
The hardest part wasn't the code — it was the design. Figuring out how to structure a 7-stage pipeline (novelty check → gap analysis → drafting → compliance audit) that actually made sense end-to-end took a long time to get right. I went through several rounds of redesigning the flow before it clicked — moving stages around, rethinking what should run in parallel versus sequentially, and figuring out where the compliance checks actually needed to sit in the pipeline.
That one design problem taught me more about system architecture than writing the actual features did.
Top comments (1)
The TF-IDF fallback keeps the workflow moving, but I'd want that mode visible all the way into gap analysis and manuscript drafting. With five academic APIs feeding the novelty check, preserving which sources responded and which papers support each proposed gap would give researchers something concrete to review. Your four compliance guards address submission requirements; carrying the evidence and its limitations between stages would help prevent a tentative search result from becoming a confident novelty claim.