Altair – Autonomous Documentation Synthesis Engine

Jun 2026 · Second place · Google Developer Groups Cloud Boston × Northeastern University Hackathon

Photos

  • The Altair team demoing the project on a laptop to a judge
  • The team walking a judge through Altair during the hackathon
  • Winning teams receiving their prizes on stage

Overview

Documentation debt is a systemic failure mode in software engineering: commit histories degrade into noise, changelogs drift out of sync, and context is lost at every handoff. Existing AI tooling forces a trade-off between routing proprietary code through cloud APIs and going without. Altair is an autonomous documentation synthesis engine that compiles version-control artifacts into deterministic, schema-validated documentation, generated entirely on local infrastructure.

Highlights

  • Engineered a multi-pass code intelligence pipeline that reconstructs documentation directly from raw git diffs, in a strict five-hour sprint under the theme “AI-First Developer Efficiencies”
  • Diffs are pre-analyzed, semantically chunked, and routed through Google's Gemma via Ollama, with structural validation guaranteeing well-formed output
  • Produces spec-compliant Conventional Commits, release-ready changelogs, and review-grade PR narratives, running entirely on-device with zero data egress

Built with

  • Python
  • Gemma
  • Ollama
  • Git
  • Conventional Commits

Team

  • Kshitij Kumar
  • Kunal Gupta
  • Jyothssena Gomatum Sreenivaasan
  • Tanmayi Shurpali
  • Stephy Romichan
View on GitHub