Syuja Dewa Kusuma@whard220599

Every plant with rotating equipment runs operator rounds: someone walks the site each shift, reads gauges on pumps, blowers and motors, and writes the numbers on a clipboard. They wear gloves, it is loud, and typing is not an option. The data reaches a system hours later, so a reading already past its alarm limit gets noticed after the bearing has failed. RONDA replaces the clipboard with a conversation. The operator names an asset, the agent loads its check points, and the operator speaks the numbers. Each reading is logged, checked against that equipment's limits, and read back aloud. On a breach the agent quotes the limit, recalls last shift's value, and offers to raise a work notification, which is sent to Telegram. A supervisor can follow the round live, and the shift report writes itself at the end. Key design rule: the model never decides whether a reading is safe. The server computes ok/warn/alarm and the agent only reads the verdict out, so the SQLite audit trail stays authoritative. Corrections supersede rather than overwrite, and ambiguous check names are refused instead of guessed. Built on the AssemblyAI Voice Agent API: one WebSocket for speech in and out, turn detection tuned for dictation, keyterms for asset tags, and four HTTP tools AssemblyAI calls directly. The report is generated through the AssemblyAI LLM Gateway. Plain Node.js, zero npm dependencies, 40 tests.

What if anyone could build a custom AI classifier in minutes, not weeks? Classifi transforms as few as five labeled examples per category into a working text or image classifier, with no model training, complex pipelines, or machine-learning expertise required. Users simply define their categories, add examples, and classify new content instantly. Every prediction includes a confidence score and a plain-English explanation, turning AI from a black box into a tool people can understand and trust. Under the hood, Classifi converts user examples into structured few-shot prompts for multimodal Gemma 3, served through Fireworks AI on AMD Instinct GPU infrastructure. This enables fast, scalable inference without requiring users to manage the underlying infrastructure. Once created, each classifier can also be exposed through a reusable REST API endpoint and integrated into real applications and workflows. From teachers assessing assignments and manufacturers detecting defects to researchers organizing data and NGOs sorting multilingual field reports, Classifi puts practical, explainable AI in the hands of the people who understand the problem best. Open-source, self-hostable, and training-free, Classifi is not merely a prototype, it is a faster, more accessible way to transform domain knowledge into specialized AI.

NexusCore is the governance brain and emergency brake that sits above AI agent teams. Built for the Band of Agents Hackathon, it addresses a growing enterprise risk: AI agents are no longer only suggesting code. They can generate patches, call tools, modify databases, deploy services, and trigger live actions faster than humans can review them. NexusCore adds governance before execution. When an agent proposes or attempts an action, the system classifies it into LOW, MEDIUM, or CRITICAL risk. LOW actions are allowed and logged. MEDIUM actions are held for review. CRITICAL actions are stopped and require explicit human confirmation before they can proceed. The system uses nine specialized agents collaborating through Band: Engineer/Builder, Proposer, Risk, Compliance, Security, Test, Infrastructure, Rollback/Audit, and Master. The Proposer Agent turns risky work into a formal proposal. Reviewer agents assess blast radius, reversibility, security, policy, test readiness, infrastructure impact, rollback plans, backups, and audit evidence. The Master Agent reads all findings and issues the final ALLOW or BLOCK decision. Band is central to the workflow, not just a notification layer. It acts as the shared collaboration room where agents exchange context, post reviews, coordinate decisions, and create a visible reasoning trail. The NexusCore dashboard mirrors this process with live workflow status, agent architecture, runtime interception, pending approvals, and an audit ledger. NexusCore demonstrates a practical enterprise use case for collaborative agents: making autonomous AI workflows safer, traceable, and human-governed before they touch production systems.

SignalScout AI is an evidence-first why-now engine for GTM and sales teams. Sales reps often spend hours researching accounts, but still reach out too late, with generic messaging and black-box intent scores. SignalScout answers one practical question: why should we contact this company right now? The system uses Bright Data as the live public-web data layer. SERP API collects fresh news, funding, product, and competitor signals. Web Unlocker extracts full text from hard-to-access pages. Web Scraper API provides pre-warmed hiring snapshots for directional hiring evidence. The backend then converts these signals into typed evidence rows with source URLs, source tiers, confidence, mode labels, and a reproducibility hash. Unlike typical AI sales tools, SignalScout does not let the LLM generate numeric scores. Scores are computed deterministically in Python from explicit signal weights, impact, confidence boost, and mode multiplier. Claude Haiku is used only for evidence-grounded synthesis: executive summary, why-now reason, sales angles, cold email, LinkedIn message, and discovery questions. The product is deployed as a live cockpit on Vercel with a FastAPI backend on Google Cloud Run. Each run shows the agent pipeline, evidence ledger, scoring audit trail, Bright Data infrastructure usage, and action pack. The result is a sales-ready GTM brief that is live, deterministic, and traceable.

Six Matches. One leaderboard. Build on AMD, earn points and compete for prizes.

🌎 Hybrid Hackathon | 💻 Online Build: 12-17 October 2026 | On-site phase: 17-18 October 2026 To participate in the hackathon, simply click Sign up with AMD. If you are not already a member of the AMD AI Developer Program (ADP), you will need to create an account. The AMD Developer Hackathon: ACT III is a hands-on event for developers, founders, engineers and builders who want to push what’s possible with AI on real infrastructure. 📍 If approved for the on-site experience, you'll also have the opportunity to showcase your project live. Travel and accommodation will not be covered.

💻 Online Hackathon | ⏳ Build: 48 hours | 📅 September 25–27, 2026 The official IBM Bob 2.0 hackathon. Build what's next in AI-assisted development, alongside the developers and experts shaping it. ⭐ Get help from experts throughout the event 🤝 Join a team or build solo 💰 Prize pool: $12,000 🧑🏻‍💻 Register before the Kick-Off Stream to start building with Bob

🌎 Online Hackathon | 💻 Month-long challenge | Dates: Sep 1–30, 2026 | 🏆 $10,000 Prize Pool ($5k cash + $5k in AAI credits) A month-long online challenge run by lablab.ai together with AssemblyAI, the Voice AI infrastructure company for builders. Every participant builds on AssemblyAI. You can join at any point — registration stays open for the whole build window, so you can start on day one or pick the project up halfway through and still submit. 🌍 Fully online hackathon. Join and build from anywhere in the world.