The problem
Employees waste hours hunting through scattered policy documents, building the same reports by hand, and switching between internal tools. Knowledge lives in PDFs, wikis, and people's heads — and none of it is one question away.
What I built
Savvy Agent is an internal assistant that gives every employee a single conversational entry point to company knowledge and tooling:
- Policy Q&A via RAG — grounded answers with citations to the source document, so people trust the response.
- Business report generation — turns natural-language requests into structured reports pulled from internal data.
- Tool actions — talks directly to the timesheet system to log, query, and summarise hours.
Architecture
- Orchestration with LangGraph — an explicit state graph routes each request between retrieval, tool-calling, and generation nodes, so behaviour is debuggable instead of a black box.
- Retrieval over a Milvus vector store, with chunking and metadata filtering tuned for policy documents.
- Serving through a FastAPI backend exposing a streaming chat API.
- Guardrails on tool actions so the agent confirms before any state-changing operation.
Outcome
A single assistant that replaces "ask around and dig through the wiki" with grounded, cited answers — and removes repetitive report and timesheet busywork.
What you get
If you need an internal copilot that actually plugs into your data and tools (not just a chatbot), I can design the retrieval layer, the agent graph, and the safe tool-calling around your stack.