Skip to content
Live

Enterprise AI HR assistant for an HRIS platform

Founder's work: a self-hosted AI assistant that answers employees' HR and policy questions and handles everyday tasks such as applying for leave, built on multi-retriever RAG and open-weight Qwen models.

HR technology

The challenge

HR teams on the platform answered the same questions all day: how many leave days are left, what a policy says, who someone reports to. Employees waited for answers that already existed in policy documents and HR records.

The assistant had to be accurate, because HR answers affect people's pay and time off. It also had to keep sensitive employee and company data inside the organisation's own infrastructure, so sending it to a public AI service was not an option.

Our solution

This project was delivered by our founder, Brayan K. Jayawardhana, outside Astrynox Labs, between June and September 2026. Brayan designed and built the whole system: the chat interface, the backend services and the AI pipeline.

Answers are grounded in the company's own policies and HR data through a multi-retriever RAG pipeline. Several retrieval strategies search a Qdrant vector store and the structured HR records, and their results are combined before the model writes an answer.

Open-weight Qwen models are served with vLLM on the organisation's own infrastructure, so no employee data leaves it. Beyond answering questions, the assistant carries out common HR tasks for the signed-in employee, such as applying for leave, checking leave balances and looking up the reporting structure.

Key features

  • Answers any HR question in plain language
  • Policy answers grounded in the company's own documents
  • Apply for leave from the chat
  • Leave balance checks
  • Reporting structure lookups
  • Multi-retriever RAG over a Qdrant vector store and HR data
  • Self-hosted Qwen models served with vLLM
  • Redis for fast sessions and caching
  • Next.js chat interface with a FastAPI backend

Results

  • Employees get HR answers instantly instead of waiting for the HR team
  • Routine requests like leave applications no longer need a manual back-and-forth
  • Sensitive HR data stays inside the organisation's own infrastructure
  • Designed and built solo in four months (June to September 2026)

Technologies

ReactNext.jsPythonFastAPIPostgreSQLRedisQdrantvLLMQwen

Ready to navigate the future?

Tell us where you are today. We'll map where technology can take you.