AgenticFinLab
Financial AI · Multi-Agent Systems · Foundation Models
AgenticFinLab develops autonomous agentic frameworks, large-model reasoning, and financial intelligence. Core focuses: multi-agent event reconstruction, visual financial understanding, emergent agent behaviors and latent planning.
Latest Updates
- May 2026Multi-Agent Entropy paper released on arXiv (2602.04234) with full code repository and simulation suite.
- Apr 2026FinMycelium v1.0: large-model multi-agent platform for event reconstruction now public. Demo and benchmarks available.
- Feb 2026PyFi (Pyramid-like Financial Image Understanding) accepted at peer-reviewed agentic workshop; adversarial agents for VLM financial comprehension.
- Dec 2025Official launch of AgenticFinLab GitHub and HuggingFace presence, open-sourcing LMBase and latent-planning codebases.
- Oct 2025iCLP: Implicit Cognition Latent Planning for LLM reasoning — preprint and code released under Apache 2.0.
All updates synchronized with GitHub releases and arXiv preprints.
Research Thrusts
Multi-Agent Event Reconstruction
FinMycelium: large model based multi-agent platform for reconstructing complex financial events, enabling traceable and explainable agent collaboration.
Visual Financial Understanding
PyFi: Pyramid-like financial image understanding for VLMs via adversarial agents. Improves chart, document and regulatory report comprehension.
Implicit Cognition & Planning
iCLP: Large language model reasoning with implicit cognition latent planning. Structured reasoning for multi-step financial inference.
Multi-Agent Entropy & Emergence
Formalizing multi-agent entropy dynamics: theoretical foundations and applications in financial agentic systems (arXiv:2602.04234).
Selected Publications & Preprints
- Multi-Agent Entropy: Theory and Emergent Coordination in Financial Agentic Systems
- PyFi: Toward Pyramid-like Financial Image Understanding for VLMs via Adversarial Agents
- iCLP: Large Language Model Reasoning with Implicit Cognition Latent Planning
- Reasoning Autoregressive Modeling for Financial Multi-Agent Inference
Full list at github.com/AgenticFinLab and arXiv.
Core Research Team
Dr. Yujie Lin
Principal Investigator
AI agents, foundation models, financial reasoning. Leads AgenticFinLab.
Dr. Chen Zhang
Co-PI / Multi-Agent Systems
Multi-agent coordination, emergent intelligence, game-theoretic finance.
Yuxin Wei
PhD Research Fellow
VLM reasoning, adversarial robustness, financial image understanding (PyFi lead).
Jiahao Li
PhD Candidate
Latent planning, LLM reasoning, code & infrastructure for agentic evaluation.
Mingqi Zhao
Research Engineer
FinMycelium core contributor, multi-agent platform engineering.
Rui Su
Graduate Researcher
Agentic evaluation, entropy dynamics and large-scale simulation frameworks.
Open positions: PhD interns and visiting scholars. Contact via GitHub/HuggingFace.
Active Repositories
Large model based multi-agent platform for event reconstruction. Orchestration, memory & reasoning.
Pyramid-like financial image understanding for VLMs via adversarial agents.
Code for "Multi-Agent Entropy: Theory and Emergent Coordination" (arXiv:2602.04234).
Supporting large model basic research: training utilities, evaluation harnesses.
iCLP: LLM reasoning with implicit cognition latent planning.
Reasoning autoregressive modeling for agentic inference.
Models & Demo Hub
Multi-agent event reconstruction model checkpoint. Supports orchestration, memory-augmented reasoning for financial forensics.
🤗 HuggingFace
agentic
Pyramid-like financial image understanding adapter for VLMs. Improves chart, table and document comprehension via adversarial tuning.
🤗 HuggingFace
VLM adapter
Simulation environment for multi-agent entropy dynamics (arXiv:2602.04234). Includes configurable agent societies and entropy metrics.
🤗 HuggingFace
simulator
Foundational large-model utilities: evaluation harnesses, agentic prompts and benchmark datasets.
🤗 HuggingFace
benchmark
Connect & Correspondence
AgenticFinLab – advancing agentic foundation models for computational finance.