MARIF: A Multi-Agent Regime Intelligence Framework for Adaptive Options Strategy Selection in Indian Derivatives Markets

Can LLM Agents Price Competitively? A Dynamic Multi-Attribute Auction Benchmark for Agentic Commerce

Agentic commerce is moving from concept to deployed infrastructure: payment networks, retailers, and AI platforms are setting the stage for agents to transact on behalf of merchants and consumers. Yet whether the LLMs behind these agents can price competently in real markets, where customer preferences are hidden, competitors adapt in real time, and demand can shift without warning, has not been systematically tested. We introduce Bazaar, a dynamic sealed-bid benchmark for multi-attribute auction under these conditions. Despite its dynamics, the benchmark is grounded in closed-form customer utilities, enabling exact evaluation. Across 11 frontier LLMs from four providers, the leading agents on customer acquisition (e.g. Gemini 3.1 Pro) are often not the leading agents on profit (e.g. Opus 4.6). The ranking shifts again under demand shocks: agents that learned fastest pre-shock are typically the slowest to revise their beliefs afterwards, while Gemini 3.1 Pro recovers fastest despite not leading on profit. However, even the strongest agent captures less than a third of hindsight-optimal profit, suggesting current LLMs are progressing in agentic commerce but leave substantial headroom.

AUTOMATING PORTFOLIO MANAGEMENT USING MULTI-AGENT SYSTEM WITH DYNAMIC PROMPT OPTIMISATION AND FEEDBACK LOOPS

A closed-loop multi-agent decision framework that introduces prompt-level learning as a scalable alternative to full model retraining and highlights the potential of prompt-level adaptation for building robust and autonomous financial decision systems.

Spillover-based information set selection: evidence from oil spot price forecasting

Mergers and Acquisitions Pricing Multi-agent Cross-Questioning: Paradigm Evolution and Framework

This paper systematically reviews the evolution direction of financial artificial intelligence (AI) from traditional statistical fitting to dynamic confrontation. With the help of functional isolation, evidence-based game theory and parameter correction, this paper analyzes the internal law of the multi-agent cross-questioning mechanism, and summarizes a set of structured analysis frameworks. Relevant studies have shown that compared with a single model, the core purpose of building a specific questioning mechanism using confrontation topology is to get through the logic of transforming unstructured semantic risk into structured valuation parameters, to clearly depict the cross-modal mapping path between the two. This model has significant theoretical feasibility in improving the internal consistency of pricing logic and restraining unreasonable premiums. This paper believes that the dynamic inquiry paradigm is conducive to optimizing and reshaping the existing asset valuation benchmark in a complex trading environment. To truly land the man-machine collaborative pricing system, the core is to focus on how to resolve the technical barriers of reasoning interpretability, privacy protection, collaborative deduction and penetration algorithm audit. Only by achieving substantial breakthroughs in these bottlenecks can the follow-up system construction have a solid foundation.

A Multi-agent Approach to Technology Intelligence: From Patents to Market Signals with Large Language Models

Technology Intelligence (TI) draws on knowledge management and competitive intelligence to give decision-makers awareness of technological trends, opportunities and threats. Recent frontier large language models and autonomous agents make it feasible to automate the full TI lifecycle. This paper presents TechIntel-MAS, an eight-role multi-agent system implemented on the CrewAI framework, that transforms a body of disclosed patent filings, supplied alongside a free-text intelligence request, into a structured competitive-intelligence report. The system treats patents as seed signals and searches the open web for corresponding market evidence (products, vendors, deployments, partnerships), and reports the absence of evidence honestly when none is found. Six frontier models (GPT-5.4, Claude Sonnet 4.6, Gemini 3.1 Pro, DeepSeek R1, Moonshot Kimi K2 Thinking, Qwen3 Max Thinking) execute the generative roles in parallel. An independent evaluator pool (Grok 4 and Llama 4 Maverick) operates the Verifier, Telemetry and Judge roles, preserving the principle that the model evaluating an artefact is never the one that produced it. We evaluate the pipeline on four heterogeneous patent filings and report per-model precision, recall, F1, semantic alignment, KIQ-relevancy and a coarse expected calibration error, together with cost and elapsed-time figures from a fully instrumented run. We make three contributions: a TI-specific eight-role architecture aligned with the Kerr three-tier model; a patent-as-seed-signal retrieval pattern that grounds disclosed inventions in open-web market evidence; and an evaluation methodology with strict generator-evaluator independence, evaluated across six frontier LLMs in a pilot study. The paper situates these results within the knowledge-management literature and discusses implications for organisational learning and strategic forecasting in agentic settings.

Concentrated Liquidity Provision: a Reinforcement Learning Perspective

Automated market makers (AMMs) are a cornerstone of decentralised finance (DeFi). Constant product markets with concentrated liquidity, such as UniswapV3, are now a well-established design. In these markets, liquidity providers (LPs) face a sequential decision problem: they must decide when to rebalance their positions and which price ranges to allocate capital to as market conditions evolve. We formulate dynamic liquidity provision as a stochastic impulse control problem and use reinforcement learning (RL) to solve it, focusing on providing interpretable solutions. We show that learned policies exhibit rich state-dependent behaviour, allocating liquidity according to mispricing, rebalancing costs, uncertainty, inventory exposure, and heterogeneous risk preferences. These behaviours help compress the left tail of the Profit and Loss (PnL) distribution and avoid catastrophic outcomes under high uncertainty. Finally, we benchmark the RL agents against baseline and sophisticated agents from the AMM microstructure literature and analyse their performance.