Rishabh Kumar
Building AI systems that run in production.
I build AI systems that run in production. Currently building autonomous email VAs processing real dealership workflows with sub-500ms hybrid RAG, real-time voice pipelines with Deepgram/ElevenLabs, and multi-tenant SaaS on AWS/GCP. I ship latency-sensitive AI systems from backend inference to production deployment.
Projects
Visual AI workflow builder. Drag-and-drop 13 node types to compose multi-step AI pipelines without writing code.
Open source NL-to-SQL library. Natural language database queries across MySQL, PostgreSQL, MongoDB with semantic caching. 40% inference cost reduction.
Multi-tenant TTS platform. Custom voice libraries across 13 categories, real-time waveform visualization, usage-based billing.
Fine-tuned BERT on MITRE ATT&CK at 89% accuracy. GNN for attack path prediction. Deployed on AWS with auto-scaling.
Multilingual event booking assistant. GPT-4o via LangChain and LangGraph with dynamic DB tools and Twilio SMS. Smart India Hackathon.
Multi-agent system for Gmail and Google Calendar. Handles chained commands like 'find a meeting time in this email and schedule it.'
Context-aware e-commerce with AI command interpretation. Cross-tab state sync via dual-layer persistence.
Experience
AI Engineer, Chennai
Built Milo, a production email VA for UK automotive dealerships. Autonomous email processing with sub-500ms hybrid RAG. Multi-tenant Qdrant knowledge base isolation.
AI Intern, Bengaluru
Multi-agent voice AI system orchestrating 5 specialized agents. Real-time Deepgram STT + ElevenLabs TTS on Plivo/Twilio. Qdrant with 1M+ embeddings at 98% retrieval accuracy.
Full Stack Engineer (ML Systems), Los Angeles, USA
Recommendation engine with transformer embeddings. Real-time inference API serving 1M+ predictions daily at 45ms P50. Apache Airflow ML pipeline on 4 GPUs.
Skills
Education
Bachelor of Technology in Computer Science and Engineering