Multi-turn Financial Advisory Dialogue on FAR-Trans 50 persona-expressive investors (test)
66.5Turn 3 Pairwise Win RateMatched Persona (Fund2Persona)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Matched Persona (Fund2Persona)Generator=Gemini 3.1 Flash-Lite, Judge=GPT-5.4 Mini, Persona Retrieval=LLM-based agentic search2026.06 | 66.5 | 78 | 3.6 | 3.26 | |
| Generic LLMGenerator=Gemini 3.1 Flash-Lite, Judge=GPT-5.4 Mini2026.06 | 49 | 31 | 3.34 | 3.14 | |
| Mismatched PersonaGenerator=Gemini 3.1 Flash-Lite, Judge=GPT-5.4 Mini2026.06 | 18 | 13 | 2.46 | 2.4 |