README.md

metadata

title: RegTech BR — Model Comparison
emoji: ⚖
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: 4.44.0
app_file: app.py
pinned: false
license: cc-by-4.0
hardware: zero-gpu

RegTech BR — Model Comparison

Side-by-side comparison of two compliance reasoning systems on the same query and RAG context:

	Model A	Model B
System	RAG + Claude Sonnet	Fine-tuned Mixtral-8x7B
Adapter	—	AutoScientist LoRA (r=32, alpha=64)
Dataset	—	adaption-brazilian-crypto-compliance (19 examples, Grade A)
Win rate	—	63% vs 37% base model

Setup

1. Upload chunks_meta.jsonl, embeddings.npy, faiss_index.bin to Space root
2. Add ANTHROPIC_API_KEY as Space Secret
3. Hardware: ZeroGPU (HF PRO account required for Mixtral inference)

Links

• Main Space (RAG only): Fernandosr85/regtech-br
• LoRA adapter: Fernandosr85/regtech-br-legal-adapter
• Dataset: Fernandosr85/adaption-brazilian-crypto-compliance
• Challenge: AutoScientist Challenge 2026