Intelligent Search

[DELIVERABLES]

PLATFORM

[IMPACT]

Knowledge Fusion

[ROLE]

Solution Architect

[SERVICES]

RAG • OBSERVABILITY • AI

[THE STORY]

This project started as a practical need: make scattered documents (e.g. manuals, LLDs , reports, etc.) searchable and usable during day-to-day work.

Solution is intelliSearch, ahybrid retrieval flow (BM25 + embeddings) with clean citations, and a structure that can scale from personal docs to team knowledge bases.

The result is a fast, explainable search experience that feels like an internal assistant—grounded, auditable, and easy to extend.

[SOLUTION]

A hybrid RAG platform combining structured chunking, fast retrieval, and citation-ready answers—optimized for reliability and scale.

GITHUB

[ARCHITECTURE]

Python

Docker

TypeScript

LocalAI

Vector store

Microservices

[DEVELOPMENT SPECIFICATIONS]

CHUNK SIZE

500 tokens

OVERLAP

120 tokens

CHUNK STRATEGY

Fixed-size

VECTOR STORE

Chroma (HNSW)

KEYWORD SEARCH

BM25 (OpenSearch)

RERANKING

Hybrid scoring (BM25 + cosine)

RETREVAL

Neighbours Aware

[USER FLOW]

User submits question → Hybrid retrieval (BM25 + vector search) → Context building → LLM generates answer → Answer returned with citations

[DETAILS]

Upload / ingest documents
Chunking + embeddings + indexing
Hybrid retrieval (BM25 + vector search)
Rerank + build context + generate answer

[WORKING DEMO]

Short walkthrough showing ingestion → search → cited answers in real-time.