INDRA – Industrial AI Copilot

INDRA is an AI-powered industrial knowledge copilot that transforms technical documents into searchable knowledge using RAG, Pinecone, Neo4j, and LLMs for intelligent troubleshooting

Comment

Inspiration

Industrial companies have thousands of manuals, maintenance procedures, spare-parts catalogs, and inspection reports. Finding the right information often takes significant time and depends on experienced engineers. We wanted to build an AI assistant that understands these documents instead of simply searching them.

INDRA was created to help engineers retrieve technical knowledge quickly using Retrieval-Augmented Generation (RAG), semantic search, and knowledge graphs.

What it does

INDRA allows users to:

Upload technical PDFs and industrial manuals Automatically extract document content Build semantic embeddings using Sentence Transformers Store vectors in Pinecone Create a knowledge graph in Neo4j Perform semantic search across documents Ask engineering questions in natural language Generate AI-assisted maintenance and troubleshooting responses

How we built it

Backend:

FastAPI Python

Database & Storage:

Supabase PostgreSQL Supabase Storage

AI Stack:

Sentence Transformers (all-MiniLM-L6-v2) Pinecone Vector Database Neo4j Knowledge Graph Groq Llama 3.1

Document Processing:

PyMuPDF pdfplumber Pillow python-docx openpyxl

The pipeline is:

Upload Document ↓ Supabase Storage ↓ Text Extraction ↓ Chunking ↓ Embedding Generation ↓ Pinecone ↓ Entity Extraction ↓ Neo4j Graph ↓ Hybrid Retrieval ↓ LLM Response

Challenges we ran into

The biggest challenges included:

Designing an efficient document-processing pipeline. Building a hybrid retrieval architecture combining vector search and graph databases. Handling large technical documents while preserving context. Integrating Supabase, Pinecone, Neo4j, and Groq into one workflow. Optimizing embedding generation and chunking.

Accomplishments that we're proud of

We're proud of creating a scalable pipeline that processes documents, performs semantic search, and generates intelligent engineering insights

What we learned

During development we learned:

Production-ready FastAPI architecture Retrieval-Augmented Generation Vector databases Knowledge graphs Semantic search Document parsing pipelines LLM orchestration API design Industrial AI workflows

What's next for INDRA – Industrial AI Copilot

Future improvements include:

Agentic multi-step reasoning Multi-document conversations Real-time industrial monitoring Predictive maintenance recommendations Multi-modal document understanding Interactive knowledge graph visualization

Built With

  • codex
  • fastapi
  • groq
  • huggingface
  • knowledgegraph
  • llama
  • llm
  • machine-learning
  • neo4j
  • openai
  • openpyxl
  • pdfplumber
  • pillow
  • pinecone
  • pymupdf
  • python
  • python-docx
  • rag
  • rest
  • supabase
  • vectordatabase
Share this project:

Updates