Devpost
Participate in our public hackathons
Devpost for Teams
Access your company's private hackathons
Grow your developer ecosystem and promote your platform
Drive innovation, collaboration, and retention within your organization
By use case
Blog
Insights into hackathon planning and participation
Customer stories
Inspiration from peers and other industry leaders
Planning guides
Best practices for planning online and in-person hackathons
Webinars & events
Upcoming events and on-demand recordings
Help desk
Common questions and support documentation
Sort by:
A safety-gated five-agent system that autonomously proposes, runs, evaluates, and remembers recommender experiments—without exposing the hidden test set.
0.9746 on the official evaluator, running offline with zero tokens. A second adversarial benchmark revealed the same agent scores 0.348 against a customer who paraphrases. Both numbers reported.
A conversational shopping agent that knows when to ask and when to answer, locating your exact product in ~4 turns with zero hosted-LLM cost.
Pramit turns uncertain R&D into predictable progress by making experiments reproducible, comparisons valid, decisions traceable, and claims defensible
A Transformer layer made 6.54x faster with every optimization measured and every dead end recorded
A local AI shopping copilot with a versioned intent ledger that removes outdated preferences and adapts recommendations when shoppers change their minds.
A shopping copilot that is stress-tested against real language (not just catalog terms), learns which questions to ask to narrow the search fastest, and diversifies recommendations for fairer coverage
An autonomous ML research agent that finds weaknesses in a recommender system, forms hypotheses, runs controlled experiments, and learns from both successful and failed results.
Adaptive verification and automatic rollback for coding agents: SafeCommit checkpoints changes, scores risk, independently verifies them, and restores the workspace when unsafe.
We built a conversational shopping copilot that understands changing intent and constraints, achieving 100% Hit Rate@10 and a 0.965 technical score, about 9x the baseline.
Git caught code collisions. Humans caught design collisions through communication. AI agents have neither. Shepherd detects when individually correct agents build a collectively broken system.
A shopping agent that treats changing your mind as a first-class event. Finds a shopper's exact product in under 3 turns even when the network dies.
Fewer questions, better matches. Turning browsing into a conversation.
“AI-powered farming, smarter decisions, better harvests.”
CropSentinel is a fortified agentic AI fleet that arbitrates satellite and ground data to execute instant, zero-trust parametric agricultural insurance payouts on Google Cloud.
A conversational shopping agent that remembers what users want, reject, and change—then combines lexical and semantic search to find the right product within ten turns.
State-bound release control for autonomous agents: certify candidates by diffing real effects, then gate promotion with an audited, tamper-evident ledger.
AI-powered disaster response that turns a citizen's SOS into a scored, prioritized, routed rescue — in real time, end to end.
A multi-turn conversational shopping agent that finds a customer's hidden target product inside a 50,000-item Amazon catalog, scoring 8.35× the official baseline with zero LLM calls or network access.
A detector is only worth its accuracy on generators it has never seen. 0.958–0.989 balanced accuracy on an unseen generator across five transforms — and we found the 2026 models it still misses.
VeriSight AI is a multimodal verification engine that scans text, images, and documents to instantly detect deepfakes, AI-manipulated media, and viral misinformation with high-precision trust scoring.
An autonomous ML research agent that builds and improves recommender models with best-first search, turning routine experimentation into a reproducible, clear workflow with minimal human intervention.
An optimized transformer layer verified on two real GPUs — three optimizations that looked safe on one backend failed on the other.
Shopping copilot!
