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 conversational shopping agent that reconstructs the buyer's hidden intent card and runs retrieval as its inverse: every clue becomes a key into a precomputed index over 50,000 products.
A shopping assistant that actually listens. It asks the right questions, remembers your answers and finds what you want.
An offline-first shopping copilot that tracks intent shifts, remembers what matters, asks a useful question, and ranks candidates from 50,000 products to turn ambiguity into a confident match.
Our system detects and redacts sensitive data, blocks high-risk prompt injection before execution, securely passes approved runtime secrets to the Runner, and provides a timeline of every decision.
Automated GPU kernel research — compile, benchmark, optimize, repeat, honestly.
An Apple M5 Pro Transformer that speeds up all reference-feasible cases and makes the 100,000-token stress case bounded, runnable, and auditable.
Find hidden products in 10 turns with zero LLM calls: Bayesian inverse inference beats BM25 0.107 at 0.9802 (100% Hit@10, MRR 1.0).
Online shoppers rarely describe the perfect product in one query. They explore, add constraints, and sometimes change their minds. We built Shopping Copilot to handle exactly that.
Ultra-fast AIGC detection on the edge, accessible to all
Offline conversational product search that asks, remembers, and reranks across turns.
Find What You Mean, Not Just What You Type.
An LLM-driven research harness that edits, evaluates, and reverts a from-scratch NumPy Factorization Machine, discovering a statistically verified within-user ranking improvement on KuaiRand-Pure.
An autonomous ranking agent that tests ideas across time, rejects fragile gains, and preserves reproducible evidence.
The trace that survives. Step-level observability that stays continuous across a context compaction and a process crash — so a long-running agent has one story, not a series of restarts.
Shopping search that listens: our agent remembers evolving needs, handles changes of mind, and combines BM25 with semantic attribute ranking to understand shoppers beyond exact keywords.
maihenduo shows you what you want, even if you don't know. The agent remembers what you say through the conversation and asks the relevant questions. It also knows when you change your mind!
Tiktok TechJam Track 4, running an AI shopping agent
Conversational product discovery—fast, private, and explainable.
TechnicalScore 0.9499 offline. The default configuration runs on the Python standard library alone: no network, no credentials, no third-party package. A local model is used when one is reachable
Shopping shouldn’t start over when you change your mind. TENFOLD remembers what still matters and updates what doesn’t. TENFOLD searches 50,000 products, scoring 0.9024 vs. a 0.1067 public baseline!
ShopSense turns messy customer preferences into ranked product recommendations. It asks the right follow-up questions, handles changing intent, and finds better matches faster.
Model Soup + Aggressive Augmentation during training. That's our key to stop AI slops.
Hand-written Triton attention kernels, benchmarked against the baseline, with shape-aware dispatch choosing the fastest path per case with every optimization validated and explainable.
Every Run tells the truth.
265 – 288 of 614
