‹ All Projects
Project Thread

FinSentry-MAP — Multi-Agent Market Intelligence Pipeline with ReAct Reasoning

🐙 View GitHub

Overview

An autonomous market intelligence system designed for high-volume crypto data ingestion and real-time synthesis using a custom Reasoning-Acting (ReAct) loop built from scratch.

Tech Stack

Python (Pandas, Requests)FastAPI & UvicornGoogle Gemini 2.5 FlashSQLite3 (Relational Modeling)Docker

Features

  • Custom ReAct Loop: Strict JSON-based tool calling with autonomous error correction.
  • Star Schema Logging: Relational database (SQLite) tracking every agent "Thought" and "Action."
  • Automated Integrity: Integrity Agent validates pricing disparities and flags true market anomalies.
  • Defect Recovery: Built-in failover logic to switch to backup API providers during outages.

Key Decisions

01

Avoided heavy frameworks like LangChain to build a lightweight, custom ReAct orchestrator that minimizes latency and JSON parsing errors.

02

Adopted a Star Schema database design to log the chain-of-thought of each agent, making the system 100% auditable for financial compliance.

Learnings

→Mastered the implementation of self-correcting JSON output loops where the agent retries logic based on caught tracebacks.
→Gained deep knowledge in MLOps by designing a data lineage system for non-deterministic agentic workflows.
EmailGitHubLinkedInResume