Trivali Feature Compliance Checker

• 

  Our compliance checker in action: the LLM retrieves relevant policies and explains whether the feature passes compliance

Inspiration

As products grow more complex and global, every new feature risks violating region-specific regulations. Tracking all these rules manually is tedious, error-prone and nearly impossible at scale. We built our compliance checker to take away that uncertainty, giving teams instant clarity on potential compliance risks before features are shipped.

What it does

The Trivali Feature Compliance Checker uses AI and a knowledge base of regulations and terminology to:

• Flag whether a proposed feature might violate compliance rules
• Identify if geo-specific logic is required
• Provide clear reasoning and sources for each flag

How we built it

We combined a Retrieval-Augmented Generation (RAG) pipeline with two vector stores:

• A policy/regulation database for compliance rules
• A terminology glossary for acronyms and internal terms

We then built an agent that queries each store intelligently and integrates everything into a conversational Gradio UI

Challenges we ran into

• Preventing hallucinations when a term wasn't in our glossary
• Making terminology retrieval strict enough to avoid false matches

Accomplishments that we're proud of

• Built a working compliance checker that can flag risks with reasoning
• Successfully implemented a RAG pipeline without prior knowledge

What we learned

• RAG is a augmentation of LLMs that allow it to gather necessary context in order to answer questions

What's next for Trivali Feature Compliance Checker

• Expand the regulation database to cover more regions
• Support dashboards for compliance teams to track flagged features at scale

Built With

• azure
• chroma
• gradio
• langchain
• langsmith
• llm
• openai
• python
• rag