bsafe

Inspiration As a Supply Chain Data Engineer at Johnson & Johnson, my Sr. Manager told me building an AI data validation layer would be "too complicated" for our architecture. I came to this hackathon with a chip on my shoulder to prove them wrong and show exactly how elegant and feasible it actually is.

What it does bsafe acts as an AI "Gatekeeper" that semantically audits incoming manufacturing telemetry against corporate compliance policies before it enters internal systems. It instantly quarantines flawed data, guaranteeing that downstream dashboards run exclusively on 100% audit-ready numbers.

How we built it. We treated this like an enterprise product, starting with strict architectural design docs before moving into rapid sprint iterations. Under the hood, we built a multi-agent RAG pipeline using IBM watsonx, Langchain, ChromaDB, and Meta's Llama 3.3 70B as our deterministic auditor.

Challenges we ran into Forcing a generative LLM to act like a strict, mathematical logic gate that only outputs pure JSON required heavy prompt engineering and custom Regex extraction layers. We also had to engineer low-level Python workarounds to resolve complex SQLite environment mismatches for our Vector Database deployment.

Accomplishments that we're proud of We successfully implemented a multi-agent RAG architecture that works seamlessly on structured tabular data with almost zero human intervention. Instead of just throwing a generic error code, the system generates automated, human-readable root-cause analyses for every flagged row.

What we learned We learned that shifting data validation "left"—catching semantic errors at the exact moment of ingestion—is infinitely better than trying to clean a polluted database later. It also proved that RAG is a revolutionary upgrade for traditional data engineering pipelines, not just a tool for building chatbots.

What's next for bsafe The technology works beautifully, so our immediate next step is B2B commercialization. We plan to package this pre-ingestion firewall as an enterprise API and sell it to manufacturing companies to help them neutralize their compliance risks.

Built With

• ai
• css3
• ibm-watson
• rag
• react