Project Case Study
Agentflow AI — Visual AI Workflow Engine
A fullstack visual AI workflow automation platform executed via a 5-agent orchestration engine for DAG planning, LLM execution, grounded validation, and fault recovery.
Architecture
Next.js/React provides the visual workflow builder and execution dashboard. Express/Node.js exposes REST & webhook APIs running a 5-agent pipeline: Planner resolves DAG dependencies using Kahn's algorithm, Execution runs LLM nodes, Validation enforces grounded outputs, Recovery handles retries, and Monitoring streams Socket.IO telemetry.
Technologies
- Next.js
- React
- Node.js
- Express.js
- MongoDB
- Mongoose
- Socket.IO
- OpenRouter
- Gemini
- Nodemailer
- Kahn's Algorithm
- WebSockets
Challenges
- Designed a 5-agent execution pipeline with clear separation of concerns across planning, execution, validation, recovery, and monitoring.
- Prevented LLM hallucinations from empty or metadata-only webhook payloads using payload preprocessing and grounded validation rules.
- Implemented asynchronous webhook acknowledgement with 202 Accepted so long-running AI workflows could execute without blocking external webhook providers.
Screenshots
