The backend is multi-service FastAPI: an orchestrator for projects, RAG, blogs, SEO, marketing, scheduling, and billing, plus specialist agents for SEO/GEO, prompts, and posts. Long jobs run on Celery/Redis with Claude and Codex runners. The panel is Next.js with client and manager workspaces and Meshy as the chat operator.
I've worked almost entirely inside this product, on both the agent backend and the panel operators use. This is what I ship day to day.
Shipped across the FastAPI multi-service backend: project auth, ingestion and RAG routing, blogs, SEO, marketing workflows, scheduling, credits, billing, and activity feeds.
Built and hardened Celery/Redis autonomous execution with Claude and Codex runners, capacity leases, audit to fix-plan to schedule, and manager pause/resume/cancel.
Owned large slices of the Next.js panel rewrite: client and manager workspaces, Meshy project chat with SSE, Discover and Manage SEO, admin tooling, and the performance work that kept data-heavy polls usable as agent output grew.
Wired Search Console, Analytics, Firecrawl, CMS and GitHub publish gates, R2 storage, and review-first content release.
A durable AI coding workspace on the web. Each GitHub repo gets its own sandbox with parallel sessions: isolated git worktrees, resumable agent runs over SSE, and a real commit-to-PR workflow shared by agents and humans. Still in active development.
Self-hostable, open-source AI research and chat app in the spirit of NotebookLM: 6+ LLM providers, RAG research spaces, cross-chat memory, configurable MCP servers, web and academic search, and a versioned artifact workspace.
Summarise, verify and chat with any YouTube video using transcripts, Gemini and RAG. Featured by LangChain, with 174 GitHub stars and 100+ active users before YouTube's crawler blocking made it unsustainable.