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
A Catalogue-Grounded Shopping Copilot Powered by Intent Transparency
11.87× geometric-mean speedup across the 12 of 14 shapes with runnable official baselines.
Building on the base of the agent launchpad, we added multi-agent workflows with revision options and restarting from stop points, authentication (password-locked with recovery keywords), and themes.
RatLab Shopping Copilot is a fully local, zero-token AI search agent that uses multi-turn state tracking and hybrid sparse/dense retrieval to accurately find products without expensive LLM APIs.
This project speeds up Transformer inference on NVIDIA T4 GPUs with packed QKV, mask caching, and a fused CUDA residual-LayerNorm kernel, achieving up to 1.537x speedup while preserving accuracy.
An Agent Access Gateway that authenticates, authorizes, budgets, and audits every model and tool call an agent makes - server-side, per call, before it forwards.
A deterministic, privacy-first shopping agent that clarifies intent, remembers preferences, and finds the right product in fewer turns.
We often "approve" AI agents without seeing what they’ll really do. AirLock changes that by isolating every run, blocking dangerous changes, and promoting only what it truly reviewed.
Real shoppers reveal what they want a bit at a time. Our agent asks the right question, remembers every detail, and swaps preferences when you change your mind, 8.2x the official baseline score.
Lab-trained AI detectors break on social media. Our model is a fast, 4-tier pipeline combining metadata, CNNs, FFT math, and CLIP to catch AI fakes even after extreme TikTok compression and crops.
The fully autonomous ML research agent reproduces the KuaiRand-Pure FM baseline, then runs own experiments and converges on its own, landing +0.0042 test-primary over baseline in 7 iterations.
Hardware-aware Transformer kernel optimization that automatically matches each workload to the right GPU optimization strategy.
ShopCopilot is a shopping agent using hybrid search, constraint-aware ranking and smart questions. It handles changing intent safely and scored a technical score of 0.74—7× the Track 4 baseline.
Kopi O(1): Conversational E-Commerce Search Agent A fully offline, stateful shopping copilot that turns vague and evolving preferences into precise product recommendations from a 50,000-item catalog.
Robust AI-face detection that survives real-world image compression, blur, and cropping.
ParLLiaMent is an autonomous ML research agent that automates the full iterative experimentation loop for recommender system benchmarks.
A build-time loop that writes, proves and selects GPU kernels per shape and per card, 116 of 116 official-shape runs pass on nine GPUs, at 3.1x to 7.1x.
Frequency-Aware GRACE-fully Corrected Pixel Features for Robust AIGC Image Detection Under Real-World Transformations
With the rise of AIGC, the line between reality and fantasy is blurred. Serious implications could follow. Can we detect them accurately under different conditions? Our team seeks to answer this.
Better ranking, one experiment at a time.
Governing and coordinating Agent Launchpad to make AI execution safe, accountable, and transparent.
MOSAIC turns unreliable AI agents into a verifiable, self-healing engineering team with DAG planning, mechanical checks, adaptive routing, and a complete audit trail.
A multi-turn shopping agent that finds a hidden target product in a 50,000-item Amazon catalog within ten turns.
To better understand the behaviour of our AIGC detector, we reviewed representative failure cases across authentic and AI-generated images.
313 – 336 of 614
