Knight Agent
An internal AI assistant that answers employees' HR questions from company documents — in seconds, not tickets.
- Role
- Product design & AI engineering
- Context
- Internal tool · Knightec Group
- Year
- 2025
- Stack
- LangGraph · FastAPI · React · Python
Questions about benefits, vacation and internal policies kept landing on the HR helpdesk, even though the answers were already written down somewhere. I designed and built Knight Agent, Knightec's internal assistant, to answer them in seconds, straight from company documents.
What I built
Answers grounded in real documents
An agentic RAG pipeline in LangGraph plans the search, refines the question and checks its own answer before replying. Hybrid search — semantic plus keyword — handles both vague and very specific questions.
A knowledge base HR runs on its own
Managers upload new policies, retire outdated files and curate links and forms without calling a developer. Documents are chunked and embedded automatically (BGE-M3, tuned for Portuguese) and become searchable in minutes.
Analytics that show the gaps
Dashboards track conversation volume, response times and the most-used documents — and surface the questions the knowledge base can't answer yet.
The outcome
Employees get instant, document-backed answers, the helpdesk sees fewer repeat questions, and HR can finally see what people actually want to know.
