Kookmin Student Wins Second Place at IBM Bob Hackathon
Kim Chan-joong, an AI Department student at Kookmin University in Seoul, placed second solo at the IBM Bob Hackathon, a 48-hour global competition run by IBM and NativelyAI (lablab.ai) that drew 5,628 participants across 1,672 teams and 503 submissions. Per Aju Press and independently confirmed on lablab.ai's official results page, Kim's project Atlas finished 2nd behind Pedigree and ahead of Sandbox. Atlas renders any public GitHub repository as a zoomable city map -- directories become color-coded districts, files become size-scaled buildings, and key files appear as landmarks -- aimed at cutting the time a new developer needs to grasp an unfamiliar codebase's structure. Built with IBM Bob, an AI coding agent with full-repository context, Atlas requires no installation: paste a public repo URL and the server generates the map. It is a small but well-attended student hackathon result, not a product launch, though the spatial-map approach to repo comprehension is worth tracking in developer tooling.
What happened
Per Aju Press, Kim Chan-joong, an AI Department student at Kookmin University in Seoul, placed second at the IBM Bob Hackathon after competing solo. The placement is independently confirmed on lablab.ai (the official hackathon platform), which lists Kim's project, Atlas, second in the "Winners and Finalists" section, behind Pedigree (1st) and ahead of Sandbox (3rd).
The IBM Bob Hackathon was organized by IBM and NativelyAI on the lablab.ai platform. Per lablab.ai, the event ran in May 2026 and drew 5,628 participants across 1,672 teams, producing 503 AI applications, with a $10,000 prize pool. Note: Aju Press attributes the event to "June 11"; lablab.ai shows the hackathon ran in May with registration closing May 15 -- "June 11" may refer to the official winner announcement date or Kookmin University's own announcement date rather than the hackathon run dates.
Technical context
Per lablab.ai, IBM Bob (bob.ibm.com) is an AI-powered development agent that operates inside a developer's coding environment with full repository context, letting it reason across an entire codebase rather than answering isolated queries. The hackathon challenged participants to build tools and workflows that developers would actually use in real projects.
Aju Press and lablab.ai describe Atlas as a browser-based web application that visualizes any public GitHub repository as an interactive city map: top-level directories appear as color-coded districts, files render as buildings sized by code volume, key files are highlighted as landmarks, and the interface supports zooming, panning, hover tooltips, and full-image download. Per lablab.ai, the pitch is "Understand any codebase's architecture in 30 seconds." Atlas performs layout calculations server-side after a user pastes a public repo URL, requiring no local installation. Per Aju Press, Kim entered the 48-hour hackathon solo and built Atlas end-to-end using IBM Bob.
Aju Press quotes Kim: "I wanted to move away from the kinds of topics AI projects tend to gravitate toward and try something fresh." Kim also said narrowing the project's scope early was a key factor: "I cut back on ambition and focused on getting the core right."
For practitioners
Repository comprehension remains a persistent productivity bottleneck for engineering teams onboarding to unfamiliar codebases. Hackathon prototypes that pair agentic coding assistants with spatial visualizations represent a practical design direction: they surface architecture and hotspots faster than linear file trees, and can be built with current LLM-backed tools in a 48-hour sprint.
What to watch
Observers should look for an open-source release or public demo URL for Atlas, support for private repositories or authentication flows, benchmarked performance on large monorepos, and whether future iterations add semantic metadata overlays such as call graphs, ownership data, or test coverage. The AI developer tooling space is producing similar spatial visualization prototypes; the key differentiation will be integration depth with IDEs and CI pipelines.
Key Points
- 1Kookmin University student Kim Chan-joong placed 2nd solo among 503 entries at the IBM Bob Hackathon with his project Atlas.
- 2Atlas renders a GitHub repository as a zoomable city map so new developers can grasp codebase structure in seconds.
- 3The spatial-map pattern for repo comprehension is a design direction worth watching in AI developer tooling.
Scoring Rationale
A creative developer tool placing second among 503 submissions at a well-attended IBM hackathon (5,628 participants), independently confirmed on lablab.ai. Practitioner signal value is real but narrow: this is a student prototype, not a product release or research benchmark. Score reflects hackathon community relevance for the AI developer tools space.
Sources
Primary source and supporting public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
