
Problem: Every developer has inherited a codebase with no documentation, no architecture notes, and the original author long gone. Understanding "what does this do and why" can take days — time that's pure overhead, not progress. This is one of the most common, most expensive, and most unglamorous problems in software development, yet almost no tooling addresses it directly. Solution: ArchForge is an onboarding tool that turns any undocumented codebase into a clear, navigable knowledge base — automatically. It uses IBM Bob 2.0's subagents to investigate a repository in parallel: each subagent is assigned a functional module (e.g. authentication, data layer, core feature logic) and traces not just what the code does, but why it was likely built that way — pulling signal from code structure, commit history, comments, and open issues. Bob 2.0's agent mode then orchestrates these parallel findings into a single, coherent artifact: a module map, a data-flow diagram, a list of known risk areas and tech debt, and a "where to start" guide tailored to common developer tasks. Document understanding is used to fold in any existing (if sparse) docs, so nothing already written is duplicated or ignored. Demo target: We're demonstrating ArchForge against Fluttergram, a real open-source Flutter/Firebase Instagram clone with ~600 forks. Its own maintainer acknowledges the codebase doesn't reflect current best practices — making it a genuine, real-world example of the exact problem ArchForge solves, not a synthetic demo. Impact: ArchForge directly targets the "onboarding" pain point named in this challenge. By replacing days of manual code reading with a single automated pass, it reduces ramp-up time, cuts the errors that come from misunderstanding legacy logic, and lowers the bus-factor risk every team carries when only one person understands a given module.
27 Sep 2026