IDentiFI

Early warning signal detection

2025
R&D / PoCpocOct 10, 2025
LangChainTavilyPydanticPython

IDentiFI started as a solo proof of concept in October 2025: an early-warning system that finds business opportunities 3–6 months before they turn into public RFQs, by watching news, funding announcements, product launches, and other market signals.

What it does

The agent runs strategic search queries through Tavily, extracts key entities and signals from the results with an LLM, scores each opportunity against a configurable ideal-customer profile, and writes out a ranked Markdown report of the top matches. I built and tuned it against a real customer's ICP.

From PoC to production

I built the original PoC end to end: search orchestration, entity extraction, ICP scoring, and report generation, on LangChain and Tavily. It was later adopted into alkemy's backend as IDentiFI, a production subagent. As part of the Luna agent rebuild I contributed the subagent-architecture refactor: migrating it to persistent state (AsyncPostgresStore), adding human-in-the-loop interrupts, and making runs resumable. From there, other engineers on the team took it further: async DB offload, DynamoDB-backed storage, and production deployment.

Gopal Khadka