Remote Full-Stack Developer at Vision Gate

• Designed and implemented a unified agent-call endpoint in NestJS/TypeScript using SSE for real-time streaming, exposing a single interface to invoke any registered tool calling agent eliminating any coupling to individual agent implementation enabling rapid additions and modifications to agent system •Architected an end-to-end multi-agent system using Mastra.ai and Vercel AI SDK , enabling hierarchical agent communication and sub-agent orchestration; implemented full-stack telemetry via Braintrust and integrated robust RAG pipelines (Vector/Agentic) using PostgreSQL pgvector, semantic chunking, and OpenAI embeddings for high-fidelity context retrieval • Implemented tool calling agents whose capabilities include CRM records manipulation, virtual file system CRUD operations, Youtube search, video transcription, web search • Engineered a role-based permission system for AI agent file-system access, returning structured denial responses that specify which principal holds access and which agent to delegate to - enabling secure, self-routing multi-agent workflows • Real-time transcription desktop app (React,Electron, WebSocket): Built a system that captures system audio from desktop, processes it through a transcription service, and streams live transcripts to a web interface using WebSocket connection •Engineered, in React/Next.js, a ChatGPT-like LLM chat experience that renders rich frontend components for different tool call types and lifecycle states, giving users explainable, auditable AI interactions • Built a collaborative ProseMirror-based rich text editor where AI agents can autonomously write, edit, and stream rich content into documents alongside users, enabling human–AI co-writing workflows • Studied and implemented OpenAI Deep Research–style multi-step workflows (search → reading → note-taking → synthesis), combining tool-calling, memory, and task decomposition to support complex research tasks inside the product • Built a docker containerised Python/FastAPI transcription microservice deployed on Azure with GitHub Actions that normalizes diverse media formats, chunks and parallelizes audio processing, and integrates the AssemblyAI API to provide fast, reliable high quality video/audio transcription