AI Agent Execution Engine
activeReAct-based autonomous research, decision, and tool execution engine built for Factagora's fact-checking platform.
Product Engineer @ Factagora Apr 2026 – Present
Highlights
- · Built a ReAct-based AI agent execution engine fully automating research → decision → execution
- · Introduced LLM adapter layer enabling Claude ↔ Azure OpenAI switching
- · Designed agent memory and skills system for cross-session long-term memory
- · Built stable execution environment with async worker (Azure Queue) + Graceful Shutdown
An autonomous AI execution engine that handles the full research-to-decision pipeline without human intervention. Built on the ReAct (Reasoning + Acting) pattern.
Architecture
The engine separates session management from the action loop, enabling multiple session types (Prediction, Claim) to share a single execution core.