Enterprise Intelligence, Autonomous and Secure. Unify fragmented data silos into a single cognitive engine, and let specialised AI agents analyse enterprise data securely under strict multi-tenant RBAC.
The problem
Plain RAG chatbots answer single questions from one source and can't tell when their answer is weak. I wanted an assistant that breaks a question into steps, pulls from documents and databases safely, and reviews its answer before showing it.
Agentic orchestration
LangGraph agent graph with multi-hop query planning, tool-using agents and a reflection step that checks and improves answers. My workflow orchestration background shaped the graph design.
Model serving
Open-weight Mistral served with vLLM, with OpenAI, Claude and Gemini behind one interface for model choice and fallback.
MCP
Model Context Protocol servers, including MCP over SQL, giving agents structured, safe access to databases and APIs.
Retrieval
Hybrid dense + sparse search on Qdrant with re-ranking; automated Google Drive ingestion, chunking and embedding.
Apps
Angular web app for chat, document management, analytics and admin; Flutter mobile client.
Backend & security
FastAPI and Go microservices using DDD and hexagonal architecture on PostgreSQL and Redis; Keycloak SSO with mandatory two-factor authentication.
Operations
Docker, Terraform and GKE; GitHub Actions for build, test, LLM evaluation and deployment; Grafana monitoring.
