Zishan Khan

Engineered a real-time autonomous Voice AI platform integrating Twilio, streaming ASR, LangGraph multi-agent orchestration, and ElevenLabs neural TTS for human-like inbound and outbound calling. Built an event-driven runtime with Redis/BullMQ, WebSockets, PostgreSQL (pgvector), and persistent call state for campaigns, interruption handling, and RAG knowledge retrieval.

Engineered a production-grade AI companion platform using Cognitive Graph architecture (LangGraph) and Next.js, implementing multi-LLM fallback orchestration (Hugging Face, Azure OpenAI, Groq). Architected dual-vector Semantic Memory (RAG) using PostgreSQL with pgvector and Redis intent caching. Built a real-time Cognitive Loop pipeline via Ably WebSockets.

Implemented supervised, unsupervised, and deep learning models in PyTorch with Weights & Biases experiment tracking (loss curves, sweeps, model comparisons). Integrated HuggingFace Transformers for open-source model inference and exposed model inference via FastAPI REST endpoints consumed by an interactive Next.js frontend with client-side 1D CNN telemetry classification.

Engineered a production-grade AI learning platform using Next.js and LangChain.js to generate structured, day-wise learning paths with Zod-validated LLM outputs. Integrated Composio OAuth to write directly into users' YouTube (playlist creation), Google Drive (document upload), and Notion (page creation) with per-user credential isolation.