Mentoring built on proof. A student's verified achievements find the right research guide. Every mentoring session is recorded and tracked in about two minutes. The college gets its accreditation evidence as a by-product.
Built for the Genesis Hackathon, Campus Problems track. The concept builds on GuruSetu, an earlier prototype by members of our team; all code here is new (see Disclosure).
Status: the backend, the ML pipeline and all 16 screens are implemented and run together in Docker Compose (
docker compose up -d --build, thendocker compose exec backend python -m scripts.seed --reset). 249 backend tests pass inside the container, and the index section 7 demo scenario runs end to end. Open items are listed in docs/phases/status.md. All data is synthetic.
- Skills are self-claimed. Anyone can type "PyTorch" on a profile, and faculty find research students by word of mouth.
- Mentoring runs on paper. Each faculty mentor looks after 10–20 students and must meet them at least twice a semester, recorded in signed registers. Nobody can see which mentee hasn't been met, or what was promised last time.
- Evidence is compiled by hand. Before every NAAC cycle, departments spend weeks rebuilding mentor ratios, issues raised and resolved, and student achievements.
| Module | What happens |
|---|---|
| Verify | A student uploads a certificate. An LLM extracts the fields. Deterministic checks (duplicate hash, name match, known issuer, valid date) decide whether it is Certified or goes to a faculty review queue. |
| Match | Fit Check for every research opening gives an explained score from verified skills, plus what is proven, only claimed, or missing. Students apply; faculty see ranked applicants. |
| Mentor | The HOD allots mentors. Mentors log individual or group sessions with attendance, topics, private notes, action items and issues. Mentees confirm attendance digitally. |
| Track | The mentor home flags who needs attention, and why (not met in 45+ days, overdue action, stale issue). Mentee 360 shows skills, academics and the whole mentoring timeline on one page. |
| Evidence | HOD and IQAC dashboards show coverage, meeting compliance and issue resolution live. A one-click draft XLSX evidence pack contains aggregates only and never private notes. |
FastAPI (Python 3.12) · Neo4j 5 (graph + vector index) · Next.js 14 + Tailwind · all-MiniLM-L6-v2 embeddings · Claude (vision extraction, structured output) · Docker Compose.
The demo also runs fully offline (LLM_PROVIDER=none), using cached extractions and manual entry.
Requires Docker with Compose v2. These commands work once
backend/andfrontend/are in the repo.
cp docs/10-devops/env.example .env # set NEO4J_PASSWORD and JWT_SECRET_KEY
# add ANTHROPIC_API_KEY, or set LLM_PROVIDER=none
docker compose up -d --build
docker compose exec backend python -m scripts.seed --resetOpen http://localhost:3000. The demo accounts are synthetic, and every password is Demo@123:
| Role | |
|---|---|
| Student | aarav@demo.gurusetu |
| Faculty mentor | meera@demo.gurusetu |
| HOD | suresh@demo.gurusetu |
| IQAC | kavya@demo.gurusetu |
Run the tests with docker compose exec -e TODAY_OVERRIDE=2026-10-10 backend pytest -q. Every other command is in environments-and-deployment.md.
The specification pack is the source of truth for the build:
| Start here | |
|---|---|
| docs/00-index.md | Document map, ID registry, assumptions, end-to-end scenario |
| AGENTS.md | Rules and read order for coding agents (Claude Code, Cursor) |
| Product requirements | Requirements with acceptance criteria, non-goals, primary flow |
| Architecture | Components, decisions, project structure |
| Backend · OpenAPI · Graph schema | Data model, API, business rules |
| AI system | Extraction, verification, fit scoring, evaluation |
| Frontend & UX | Screens, flows, components |
| Security & privacy | Access rules, privacy matrix, threat model |
| Test strategy | Tests, failure modes, QA checklist |
| Phase plan · Status · Parallel work | Hour-by-hour tasks per person, gates, add-ons, freeze; file ownership for conflict-free work |
| Team plan | Demo story, disclosure, pitch material |
- Fit, never rank. Scores describe one student against one opening. There are no leaderboards and no ratings of faculty.
- Private by default. Session notes are visible only to the mentor who wrote them, and are excluded from every dashboard and export.
- Honest labels. The UI marks what is Certified versus Claimed, what is a DRAFT, and what is sample data. Accreditation criterion codes are drafts until checked against the official NAAC manual.
- Synthetic data only. No real student data is stored or demoed.
The concept and name come from GuruSetu, an earlier prototype by members of our team, exhibited at TechFair 2026. No code from that prototype is reused: the backend, frontend and data were written from scratch for Genesis. [VERIFY: whether the Genesis rules also restrict reusing an earlier project's idea; confirm and adjust this text.]
Built for Genesis:
- certificate verification and the trust ladder;
- explainable Fit Check matching;
- the digital mentor-mentee system (allotment, session logs, action items, issues, tracking);
- the coordinator dashboards and draft evidence export.
The earlier prototype's matching was keyword-based; the keyword method is used here only as the evaluation baseline.