The Problem

AI labs move fast. But there is a lag between what a lab is doing and what they announce. The hiring signal always comes first.

Every cluster of "inference optimization engineer" postings, every sudden wave of "alignment researcher" roles, every burst of "foundation model" keywords in job descriptions tells a story months before a product launches or a funding round closes. Job boards can be noisy, unstructured, and scattered across dozens of sites. Connecting a hiring surge to a research direction requires correlating job postings, arXiv papers, funding news, and domain taxonomy in a way that is impossible to do manually at scale.

HireSignal automates exactly that.

The Solution at a Glance

HireSignal is an agentic pipeline built to answer one question: who is building what, and how urgently?

Given a list of AI companies and a time window, HireSignal:

  1. Fans out N concurrent research workers, one per company
  2. Each worker pulls structured job data from the Greenhouse API, unstructured postings via Tavily web search, funding and partnership news, and recent arXiv papers (cs.AI / cs.LG / cs.CL)
  3. An LLM extracts skills, domains, role types, and team signals from raw text
  4. A deterministic scoring engine produces an explainable 0-100 AI intent score with full breakdown
  5. An LLM writes per-company intelligence narratives grounded in actual fetched data
  6. A final LLM pass synthesizes a ranked intelligence brief with cross-company patterns

The whole pipeline is backed by a FastAPI server, surfaced in a Streamlit dashboard, and built to be production ready: fault-tolerant, observable, and explainable.

1.png

2.png

Architecture

The pipeline compiles as a LangGraph StateGraph, a directed acyclic graph where each node is a stateful agent and edges encode execution order.