Prince Kumar@princekr144
3
Events attended
1
Submissions made
India
1 year of experience
About me
I'm a CS student who takes deadlines seriously — if I say I'll finish something, I finish it, and I try to do it properly, not just get it done. Right now I'm in my final years of B.Tech in Computer Science at ABV-IIITM Gwalior (graduating 2027), and lately I've been really into full-stack dev, distributed systems, and agentic AI — basically whatever lets me build something that actually works end to end. This past summer I did a Summer Colloquium as a Software Developer at IIITM, working on a real-time grid monitoring platform for SLDC Odisha. I built out an API layer, hooked up live SCADA + weather data across 5 grid zones, and worked on a load forecasting pipeline — my first time working with actual production-scale real-time data, and it taught me a lot about building things that can't just quietly break. Before that, I was Tech Lead at Sangillence, where I basically built the Open Book Olympiad platform from scratch — React frontend, interactive content, secure student registration with live photo verification. We shipped it for 500+ students in about 14 weeks, which was honestly a lot of late nights but a great experience. On the side, I build stuff I'm curious about: - Roleneo — a career transition tool that uses a 15-agent pipeline (LangGraph) + GraphRAG on Neo4j to figure out skill gaps and suggest a roadmap toward a target role - QueryForge — a distributed SQL engine I built across 3 worker nodes with gRPC, with fault tolerance and full observability (Prometheus/Grafana) - An AI blog generator that can turn voice input into multilingual blog posts using LangChain + ElevenLabs I like projects where I have to actually think about the architecture, not just write code that works for the demo. Still learning a ton, but that's kind of the point right now.
🤝 Top Collaborators
.png&w=640&q=75)
🤓 Latest Submissions
TraceMap
CRITICAL GAPS IN MODERN DEBUGGING Green test suites create false security. Our sample expense reporter crashes with a ZeroDivisionError despite 3 passing unit tests—because test data populated every expense category while real data lacked training expenses this month. Debugging forces developers to manually map caller chains, calculate risk, and construct regression tests. Current AI debuggers fail here because they guess code from raw traceback text without structural awareness or verification. THE TRACEMAP ENGINE TraceMap divides the repair process between two complementary components: 1. DETERMINISTIC INDEXING — CODEMAP Utilizing CodeMap (our pre-existing open-source static call-graph dependency), TraceMap indexes the repository and maps stack frames to exact symbols. It extracts failing source code, direct callers, depth-ranked transitive blast radius, and covering unit tests. Unresolvable frames are marked "not in index" to eliminate hallucinations. 2. REASONING & REPAIR — IBM BOB 2.0 Operating in tracemap-fixer mode, IBM Bob 2.0 evaluates the evidence pack. On our sample crash, Bob identified that while divide_total raised the exception, the root bug was one frame above in average_per_category (len([]) passed on empty records). Bob fixed the caller with a 2-line guard clause rather than masking the crash line. 3. MATHEMATICAL PROOF & VERIFICATION TraceMap rejects unverified suggestions. A fix is accepted only when: • The generated regression test FAILS on original code. • The regression test PASSES on the patch. • All pre-existing test suites remain passing. • The crashing application command exits cleanly with Code 0. WORKFLOW & TARGET AUDIENCE Targeted at Python developers working on complex, inherited codebases. Users interact through a unified CLI (tracemap fix crash.txt) or a live TUI showing real-time pipeline status. TraceMap replaces guesswork with mathematically verified bug repairs.
27 Sep 2026
