Enterprise Agentic RAG Platform
Production RAG for enterprise knowledge — ingestion, retrieval, evaluation, and multi-agent orchestration on Azure.
- Role
- AI Engineer
- Timeline
- Ongoing · Accellor
- Stack
- Python, LangChain, Azure OpenAI, PGVector, FastAPI
Problem
Enterprise knowledge was scattered across SharePoint and related stores. Teams needed grounded answers at scale without sacrificing auditability or latency.
Solution
I designed and deployed an Agentic RAG platform that ingests enterprise documents, embeds them into PostgreSQL/PGVector, and serves multi-agent retrieval workflows on Azure OpenAI with evaluation and caching.
Architecture
Document sources flow through ingestion jobs into PGVector, then retrieval and agent orchestration serve grounded answers.
- Enterprise Docs (SharePoint / Blob)
- Ingestion & Embedding Jobs
- PostgreSQL / PGVector
- Retrieval + Re-ranking
- Multi-Agent Orchestration
- Azure OpenAI Generation
- Evaluation & Caching Layer
My Contribution
Owned platform design and delivery: multi-agent orchestration, ingestion with delta/retry processing, FastAPI microservices, CI/CD, private networking, and RAG evaluation that drove latency and relevance improvements.
Technology Stack
Results / Impact
17K+
enterprise documents processed
90%
reduction in response time
Azure
private, VNET-secured deployment
Learnings
Chunking strategy, caching, and evaluation metrics moved the needle more than swapping models. Grounding and observability kept the system trustworthy for enterprise use.