HN Showcase
Back to browse
GitHub Repository

Used Perplexity computer to create agentic pipeline to predict CFPB enforcement actions

2 starsJupyter Notebook

AI agent autonomously used BoTorch to predict CFPB enforcement actions

by mshipman·Mar 16, 2026·1 point·0 comments
Visit ProjectView on HN

AI Analysis

●●SolidWizardryBold Bet

AI agent autonomously selected BoTorch and tuned hyperparameters without human intervention.

Strengths
  • •Agent independently chose library and model architecture without specific prompting.
  • •Bayesian Optimization pipeline beat random search baseline by significant margin.
Weaknesses
  • •Small test set of 16 samples risks significant model overfitting.
  • •Specific to CFPB data making it hard to generalize elsewhere.
Category
AI/ML
Target Audience

Data scientists, compliance analysts

Similar To

H2O.ai · TPOT · DataRobot

Post Description

Gave Perplexity Max a single task: predict CFPB enforcement actions from public complaint data. Told it nothing about BoTorch — it found the repo, chose MixedSingleTaskGP, ran 48 BO vs 48 random evals. BO mean F1=0.725 vs 0.389. Total cost: $200. The model is now scoring live companies. Happy to answer questions about the methodology or the BO search space design.

Similar Projects

Security●●Solid

Agent Gate – Execution authority for AI agents, vault-backed rollback

The idea of inserting a deterministic 'gate' between proposed tool calls and execution is smart and practical: precomputed classification plus an agent-unreachable vault means destructive operations become reversible by default. The README calls out clear mechanics (envelope checks, vault backup, tiered responses and structured denials) which suggests this is more than a thought experiment — it's a focused infra piece for teams that actually let agents touch production.

Big BrainNiche Gem
hammer32
113mo ago