What it does
This project is a prototype system that utilizes LLM capabilities to automatically flag features that require geo-specific compliance logic. It is designed to reduce manual effort, mitigate regulatory risk, and provide a clear, auditable trail for compliance checks.
How we built it
Python, OpenAI API (gpt-4.1-mini)
- Lightweight rules filter obvious business-only geofences.
- Small glossary/ontology adds domain hints (regs + trigger phrases).
- LLM prompt returns strict JSON with is_geo_compliance_needed, reasoning, relevant_regulation.
Files: main.py → reads CSV, calls analyzer, writes results. analyzer.py → builds the prompt, calls the model, parses JSON.
Challenges we ran into
- Starting the project as beginners using LLM for the first time.
- Selecting model to use. We went through a few free models from Ollama before deciding to invest and use a more updated model from OpenAI so that it is able to search the web for geo-specific search for regulations.
- Determining whether result produced is accurate or results forced to fit what we expect.
- Tuning prompt for precision
What we learned
- Implementing a LLM for specific use
- Finetuning of prompts for consistent results
What's next for clock llm
- Implementing UI
- Allowing upload of more artefacts (PRD, documents)
- Further finetuning
Built With
- openai
- python
Log in or sign up for Devpost to join the conversation.