🛠 Practical AI Builds: Scheduling Agents, MoE Assembly, Virtual Try-On, and More – August 11

Rafael Mendiola showcased a React web app that vibes code into polished slides, cutting the tedium of Google Sheets and Keynote. The demo highlighted a declarative vibe-code interface that auto drafts content and applies visual styling, making slide creation faster and more repeatable. A MIT-trained frontend veteran from Hinge Health, Rafael brought practical, designer-friendly AI tooling to a real-world workflow. Subtle survey feedback echoed the value of this approach, and the concept has real product potential for branded, AI-assisted decks.

Her contract analysis demo showcased a pipeline that ingests sample agreements, automatically extracts key terms, flags risky clauses, and highlights non-standard language to cut review time. It combined clause-level parsing with a Hugging Face model, plus prompt-like logic and pattern matching, surfaced in a Streamlit UI with before/after annotations. Gowsiya S. Shek, a data scientist at Ford/HTC Global, brings GenAI and cloud ML chops. Survey feedback noted its practicality, hinting at a scalable enterprise tool for contract workflows.

VibeCode: from design to generative AI showed how a natural-language idea can become a production-ready full-stack app using the 90/9/1 methodology—90% deterministic template-based code, 9% AI-driven refinement, 1% manual business tuning—across Go or NestJS backends, React with Atomic Design and GSAP frontends, and reusable Swagger-documented modules, supporting databases like PostgreSQL, MongoDB, SQLite, and Supabase. Founder Aldo Muñoz Ortiz demonstrated this still-in-progress workflow, which slashes backend build costs from $80-150 and 40-90 minutes in an LLM-only flow to $3-7 and 3 minutes, delivering consistent, scalable outputs. The demo resonated for its concrete, reproducible path to rapid prototyping and productization, offering startups speedups of up to 30× and cost savings of up to 95%.

Jared showcased an engaging demo on building a multi-agent email generation tool that achieved impressive open and reply rates using dynamic content pipelines. He drew on his strong background from PromptLayer and a solid UC Berkeley education to engineer a system that coordinated multiple agents for seamless email automation. People loved how it tackled real email challenges while hinting at a practical, revenue-generating product, a story that truly resonated with our community.

Jacob Kaempf presented Everyday, a platform that uses LLMs, MCP orchestration, and ReAct-based reasoning to automate multi-step tasks across apps for creative operations. The demo blended agentic decision making with computer vision to drive asset-driven workflows in real time. As a founder/CEO, he kept the focus practical and product-oriented. (people loved the practical automation angle). If turned into a product, it could scale cross-app creative pipelines for teams and agencies, a practical blueprint for builders to automate workflows.

Signal from the Noise showed how Gemini AI can extract structured entities from unstructured jazz blogs, with a live site to browse reviews while listening to tracks. It featured a Node.js Gemini API walkthrough, including prompt design, Gemini Flash 2.5 schema setup, and API calls, plus mapping findings to Spotify and YouTube. Mark Watkins, founder of Bookship, brought clear product focus and hands-on debugging. Audiences responded positively, and the project resonated with the move toward practical AI-powered content curation.

Pratim showcased a Rust-based CLI tool that chatted with codebases using Candle, embeddings, ANN, and LLMs, and also touched on HyDE and Reranking. She built a local search agent where developers could privately search and interact with their code, a refreshing approach emphasizing low cost and trust. Pratim, a dedicated Senior Developer Advocate, impressed all with her clear, hands-on style. People loved how the demo provided a practical tool for coding workflows and offered insights into emerging trends in local AI tooling.

It highlighted 'Building AI Agents with Google Cloud AI,' a session walking through Google's media capabilities with Imagen and Veo on Vertex AI and how to stitch them into agentic workflows. Katie Nguyen, Seattle-based Google Cloud AI Developer Relations Engineer, guided attendees with practical prompts and agent components, drawing on five years in the field. The talk earned clarity scores, with a nudge for more exploration; it aligns with the agentic AI and cloud deployment trend with value for builders.

Solomon Tsao gave a fast-paced demo on going from idea to live site with Gemini CLI and Bolt/Lovable, showing how to scaffold a personal blog, generate content and SEO hints with AI agents, all inside the lightweight vibe_coding_personal_suite repo. As a founder and software engineer from The Colony, TX, he demonstrated pragmatic automation and end-to-end tooling. The audience loved the reproducible workflow and practical takeaways; it hints at a scalable, client-ready web automation product.