Inspiration
We were inspired by a single question: what if startup creation itself could be redesigned — not as a static brainstorm, but as an AGI-powered orchestration of thought, creativity, and strategy? Too often, ideation tools are opaque or shallow. We envisioned a platform where language models, vision models, and logic chains work together visibly, transparently, and in real time — helping anyone, anywhere, spawn world-class startup blueprints instantly. The deeper motivation was to show how agents can think together like a team of cofounders — not just complete text.
What it does
StartupMesh is a multi-agent AGI ideation engine that transforms a single prompt into a startup-ready dossier, complete with:
Market analysis Action plan Creative assets Compliance briefings AI-generated product visuals All visibly orchestrated through Claude, Gemini, and Groq agents Users input any idea — from climate tech to quantum cities — and the system dynamically routes tasks to the best agent. Each step is rendered in real-time in an elegant, explainable UI powered by Streamlit.
How we built it
The frontend was built in Streamlit, customized with HTML/CSS to create a futuristic, neon-lit UX that reflects the intelligence under the hood. The backend is a modular FastAPI service that orchestrates requests between Claude 4, Gemini 2.5, and Groq — each selected for their domain strengths (reasoning, multimodal, speed). We deployed to Render.com using a GitHub-connected pipeline. API keys are securely stored, and inference is batched across agents using a custom routing layer inspired by Unify and LangGraph.
The final output is a full-stack, real-time ideation system with explainable AI steps, visual artifacts, and a downloadable startup dossier.
Challenges we ran into
Latency & rate limits between different LLMs forced us to build fallback logic and async-safe pipelines. UI complexity in Streamlit meant we had to push its layout engine beyond default settings to create custom containers, hover effects, and loading animations. Cloud deployment friction added complexity — especially when Render hit cold-start issues and GCP’s container setup got too heavy for fast iteration. Orchestrating multiple APIs with different auth models and output formats required deep abstraction and smart pre/post-processing.
Accomplishments that we're proud of
Built a working multi-agent AGI orchestration system, live in the cloud Fully visual step-by-step reasoning UI Used Claude, Gemini, and Groq together — rarely done in a single hackathon project Delivered a real startup prompt into a complete plan + product image + downloadable output Deployed on Render with a clean URL and public access under tight time pressure What we learned
We learned how to combine state-of-the-art models into a single cohesive experience that feels intelligent, visual, and creative. We mastered Streamlit's layout and routing customization, implemented real-time agent logic, and built around cold-start and rate-limiting obstacles. Most of all, we learned that great AI systems aren't just about performance — they're about clarity, explainability, and user trust. StartupMesh is more than a demo — it's a new form of cognitive collaboration between agents and humans.
What's next for StartupMesh: AGI x Y Combinator
We’re evolving StartupMesh from a demo into a full-scale AGI startup studio. Next steps include:
Deploying to GCP Cloud Run for scalable, production-grade infrastructure Adding memory + RAG, so agents can reference previous startups, pitch decks, and decisions Integrating with tools like Notion, Figma, and GitHub Copilot to output full MVPs Building a team creation agent that recruits cofounders, suggests roles, and drafts job descriptions Launching StartupMesh Pro, targeting early-stage founders, accelerators, and global solopreneurs We plan to submit StartupMesh to Y Combinator and open-source a free version for public ideation. Our belief is simple: Startup creation should be sovereign, agentic, and accessible to all.
Let me know if you want me to tighten the GitHub README or add submission tags/descriptions for visibility. You’ve got a real winner here.
Log in or sign up for Devpost to join the conversation.