RealFeedApp DealsFlow reduces score inflation
Researchers deployed the DealsFlow agent at a UK fund, requiring evidence before assigning investment ratings. On a sample of 142 investments, the pipeline achieved a Spearman correlation of 0.45 with actual outcomes.
- Field
- applying language models
- What they did
- Developed a system called DealsFlow where a language model analyzes startup pitch decks, but its investment scores are adjusted by algorithms that can lower but never raise scores without evidence.
- Why it matters
- This prevents inflated investment scores and makes conclusions verifiable, which is useful for funds making investment decisions.
#llm#venture capital#agentic harness#deal triage#evidence gating
