Fine-tuning use case: investment signal filtering — open source Claude Code skills

[tellmefrankie]

Hey LlamaFactory community,

Fine-tuning framework builders — sharing an open source project that could benefit from fine-tuning: Claude Code skills for investment analysis.

AI Investment Skills:

• Options flow scanner — sector ETF P/C anomaly detection with volume quality filtering
• Price monitor — stop-loss/take-profit Telegram alerts
• Daily AI briefing

Fine-tuning angle:
The options scanner uses a classification step: "Is this volume genuine institutional positioning, or lottery-ticket retail noise?" Right now this is done via prompt engineering (filtering $0.01 strikes). Fine-tuning a model on labeled historical options data (genuine_signal: bool, junk_pct: float) could make this classification more accurate and efficient.

Current approach:
Prompt-based filter removes: low-strike calls (<$0.05), extreme OTM, near-expiry lottery plays. After filtering: XLI P/C dropped from 5.32 (noise) to a meaningful institutional signal.

Open source:
→ https://github.com/tellmefrankie/ai-investment-skills

Anyone here fine-tuning models for financial classification tasks with LlamaFactory?