TriAgent: Divergence-Aware Multi-Agent Committees for Cost-Efficient Financial Sentiment Analysis

2026-07-22 · Source: Computation and Language · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, quick

Summary

TriAgent is a multi-agent committee designed to reduce the high costs of production LLM-based financial sentiment analysis. It stratifies agents by contextual granularity, using a word-level lexicon (VADER), a sentence-level domain transformer (FinBERT), and a cross-sentence reasoner (Qwen2.5, 0.5B-14B-4bit, with Mistral-7B and Phi-3.5-mini checks). A Semantic Divergence Index (SDI) routes queries based on pairwise disagreement. Key findings include a critic plateau where an LLM re-tasked as a critic achieves F1 ~0.87, outperforming a 3-persona vote (F1=0.66). The SDI also enables a Shared Consensus Dictionary for 95% of Chinese queries at F1=0.99 and acts as an LLM-hallucination detector (AUC=0.90). TriAgent achieved a Sharpe ratio of 3.50 in a 20-ticker back-test, significantly surpassing always-FinBERT (1.36) and always-LLM (0.11). At 10M users, it saves \$9.3M/year compared to a GPT-4o-mini baseline.

Key takeaway

For Directors of AI/ML optimizing LLM inference costs in financial sentiment analysis, you should consider implementing a stratified multi-agent committee approach like TriAgent. This method significantly reduces operational expenses, potentially saving \$9.3M/year at 10M users, while improving accuracy and providing robust hallucination detection. Evaluate integrating a Semantic Divergence Index to intelligently route queries and enhance system reliability.

Key insights

Divergence-aware multi-agent committees offer cost-efficient and robust financial sentiment analysis by stratifying contextual granularity.

Principles

• Stratify agents by contextual granularity
• Route queries using semantic divergence
• LLMs as critics outperform multi-persona voting

Method

TriAgent combines VADER, FinBERT, and Qwen2.5 (0.5B-14B-4bit) with a Semantic Divergence Index (SDI) to measure disagreement and route queries.

In practice

• Use Shared Consensus Dictionary for cross-lingual queries
• Apply SDI for post-hoc LLM hallucination detection
• Implement multi-agent committees for cost savings

Topics

• Financial Sentiment Analysis
• Multi-Agent Systems
• LLM Cost Optimization
• Semantic Divergence Index
• Qwen2.5
• Hallucination Detection

Best for: CTO, AI Architect, AI Engineer, AI Scientist, Machine Learning Engineer, Director of AI/ML

Related on AIssential

• TriAgent: Divergence-Aware Multi-Agent Committees for Cost-Efficient Financial Sentiment Analysis — Granularity-stratified multi-agent committees with divergence-aware routing enable cost-efficient, accurate financial sentiment analysis and hallucination detection.
• The Cost of Consensus: Isolated Self-Correction Prevails Over Unguided Homogeneous Multi-Agent Debate — Unguided multi-agent LLM debate is costly and often degrades accuracy due to sycophancy and contextual fragility.
• Precision over Probability: Orchestrating AI Agents for High-Stakes Finance — Decoupling LLM reasoning from deterministic calculation is crucial for accuracy in high-stakes financial AI.
• [D] Self-Promotion Thread — The AI/ML community actively develops diverse tools, from cost-efficient LLM interaction to advanced model evaluation and production monitoring.
• Quantifying the Accuracy and Cost Impact of Design Decisions in Budget-Constrained Agentic LLM Search — Budgeting search depth and retrieval quality offers more accuracy gains than expanding token limits for agentic RAG.
• How to build Multi Agents for FINANCE: Outperforming Anthropic — Specialized, locally runnable LLMs can outperform large proprietary models in domain-specific tasks through targeted training and novel evaluation.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computation and Language.