Zero Shot at Winning

Compliance Checker Page
Upload Regulations Page

Overview

Our project introduces a prototype system that uses Large Language Models (LLMs) as part of a multi-agent RAG system to automatically flag features requiring geo-specific compliance logic. The solution provides auditable outputs, enabling proactive legal guardrails and traceable evidence for regulatory audits.

Problem Statement

Detect whether a TikTok feature needs geo-specific compliance logic based on feature artifacts such as titles, descriptions, and documents (PRD, TRD).

Outputs:

Flag for geo-specific compliance requirement
Human-readable reasoning
Related regulations

Key Features

Automated Screening – Detects whether a feature requires region-specific compliance logic.
Explainable AI – Produces clear reasoning, risk level, and related regulation references.
Multi-Agent Workflow – Screening, Research, Validation, and Learning agents collaborate to improve accuracy.
Feedback Loop – Human feedback continuously enhances the system’s precision.
Document Ingestion – Upload PDFs of regulations, which are chunked, embedded, and stored for retrieval via ChromaDB.

Tech Stack

Backend: Python 3.11+, FastAPI, LangChain
LLM Integration: OpenAI GPT-4o-mini with LangGraph multi-agent orchestration
Vector Database: ChromaDB
Frontend: Streamlit
Infrastructure: Docker, Docker Compose, PostgreSQL

Data

No additional datasets were used, all data used was obtained from the problem statement provided by TikTok Techjam 2025. Regulations included are the focus of our solution, but users are able to upload new regulation documents easily.