Five-Mode Enterprise RAG Suite

Team track record (prior role) · AI / Team track record

A local-first retrieval suite with five modes: Hybrid, Graph, Multimodal, Agentic and Multi-Agent RAG.

From our team's track record: our AI lead contributed to this as part of an R&D team in a previous role. This was not a NezT Digital client engagement.

What was built

Hybrid RAG combining dense vector search with BM25.

Graph RAG extracting entities and relations into Neo4j.

Multimodal RAG using visual page embeddings, speech-to-text and text-to-speech.

Agentic RAG with a LangGraph self-correction loop, retrieval grading, web fallback and hallucination checks.

What our ai lead owned

Local model loading, pipeline integration, retrieval and reranking, and evaluation.

The agentic self-correction flow and the orchestration layer that routes queries between pipelines.

Validation against RAGAs quality metrics and latency benchmarks.

Tech stack

LangGraph, Neo4j, spaCy, BM25, ColPali, Kokoro TTS, Tavily, RAGAs