Inspiration

Inspiration

HR teams juggle a dozen disconnected tools for hiring, attendance, payroll, performance and exits. The data is scattered, and "AI" in HR usually means a black box that nobody can audit. We wanted an HR system where AI genuinely helps, but where every answer is grounded in real data and every decision stays with a human.

What it does

HIVE is an end-to-end AI-HRMS built around 17 modules:

  • Core HR: employee records, documents, org structure, job families and roles
  • Attendance & shifts: shift assignment and attendance tracking
  • Payroll: payslips, earnings and deductions, tax calculation, adjustments (bonus, loan instalment, allowances)
  • Offboarding: exit cases, task checklists, final settlement (completing a case is blocked while tasks are open)
  • Skills & gaps: skill inventory and automatic skill-gap sizing per role
  • Security: MFA, 7 roles and 12 permissions, with team-scoped access for managers
  • AI Copilot: a conversational assistant for HR questions, powered by Groq

Three principles run through the whole system:

  1. The AI never bypasses authorization. Permissions are checked on the backend before any data is retrieved, so the copilot can only see what the logged-in user is allowed to see.
  2. AI recommends, humans decide. Hiring, payroll and exit outcomes are suggestions, never automatic actions.
  3. Answers are grounded in the company's own data (RAG) rather than guessed.

How we built it

  • Backend: FastAPI + SQLAlchemy 2 on SQL Server 2022, with 18 Alembic migrations
  • Async work: Redis + Celery
  • Frontend: React 18 + TypeScript + Vite with a custom design system
  • AI: Groq (openai/gpt-oss-120b) behind a permission-aware copilot layer
  • Infra: everything runs with a single docker compose up
  • Testing: 100+ automated backend tests that run against both real SQL Server and SQLite, 50+ Playwright end-to-end tests, and a full smoke-test script. We also re-ran the suite on shifted dates to catch time-dependent bugs.
  • Data: the demo uses only public and synthetic data, generated by a seed script that fills every table

Challenges we ran into

  • Making tests behave identically on SQL Server and SQLite (computed columns, constraints, cleanup between tests)
  • Designing role-based access so a manager sees only their team while HR sees everyone, and enforcing it in the backend, not just the UI
  • Keeping the AI copilot useful without ever letting it leak data the user is not entitled to

What we learned

Good HR software is mostly about trust: permissions, audit trails and explainability matter more than a flashy model. Treating AI as an assistant on top of a solid, tested system worked much better than putting AI at the center.

What's next

  • AI-assisted hiring (keyword vs semantic resume matching with precision/recall/F1 evaluation)
  • RAG-based policy Q&A with groundedness metrics
  • Attrition prediction and workforce forecasting
  • Payroll anomaly detection
  • QR / geo-fence attendance, reimbursements and an AI meeting assistant ## What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for HIVE: AI-Powered Human Resource Management System

Built With

Share this project:

Updates

Submission history