Engineering Intelligence Platform for 300+ Engineers
Built a multi-agent platform from scratch that gave a 300-person engineering org real-time visibility into capacity, skills, and delivery health.
Challenge
Zero visibility into team capacity, skills, or delivery metrics across 300+ person org
Solution
Multi-agent platform — FastAPI, Neo4j, PostgreSQL, Qdrant
Result
Real-time capacity mapping, skill visibility, delivery analytics adopted org-wide
The Problem
A 300+ person engineering organisation operating across multiple accounts, geographies, and tech stacks had no single source of truth for capacity, skill distribution, or delivery health. Managers relied on spreadsheets that were stale before they were finished. Staffing decisions were made on gut feel. When a critical skill was needed — say, someone who knew both React Native and payments APIs — no one could answer "do we have that person?" faster than a three-day email chain.
The cost was real: misallocated engineers, duplicated effort across accounts, and delayed staffing that slowed delivery timelines by weeks.
What I Built
I designed and built — not just managed — a multi-agent Engineering Intelligence Platform. The stack:
- FastAPI for the API layer — lightweight, async, and fast enough for real-time queries
- Neo4j as the graph database — engineers, skills, projects, and allocations modelled as a connected graph. "Who knows FastAPI and is available next sprint?" became a single Cypher query
- PostgreSQL for structured delivery data — sprint velocity, burn-down, capacity utilisation
- Qdrant as the vector store — embedding engineer profiles and project descriptions for semantic matching ("find me someone like this senior who just rolled off")
The platform ingested data from Jira, Confluence, GitHub, and HR systems. A set of AI agents ran nightly to update skill graphs, flag capacity risks, and generate staffing recommendations.
The Outcome
Within 8 weeks of launch, the platform was adopted org-wide by 300+ team members. Results:
- Capacity mapping dropped from a 3-day spreadsheet exercise to a real-time dashboard query
- Skill gaps became visible before they became blockers — the graph showed exactly where the org was thin
- Staffing recommendations were generated automatically, reducing the sales-to-staffing cycle by 40%
- Leadership used the delivery analytics layer for executive reporting — replacing four different report formats with one
This wasn't a side project. It was a production platform that changed how the organisation made staffing and delivery decisions.