ComplaintHub
Voice-to-voice citizen grievance registration over live phone calls, with structured complaint data extracted into Redis and PostgreSQL.
- Deepgram
- Groq
- Sarvam TTS
- Silero VAD
- Twilio
A real-time STT → LLM → TTS platform that calls customers who abandoned a booking. It responds in under 900ms and lifted booking conversions by 10%.
A dual-chain LangChain chatbot with MCP tool calling, persistent memory and channel-aware replies for flights, hotels, trains and buses.
A 10-stage LangGraph pipeline covering 434 templates with hybrid retrieval, PII masking and a quality gate. It cut support turnaround by 30%.
I design, build and deploy production Conversational AI and Voice AI systems across Web, WhatsApp and Telephony. The backend is async-first FastAPI microservices, containerised with Docker.
Built a breast-cancer prediction model with SVM (98.25% accuracy) and a stock-movement prediction system that benchmarked 6 ML models on 5 years of NIFTY50 data.
Built a 3-class Twitter sentiment classifier in PyTorch on TF-IDF features, and fine-tuned ResNet50 on Food-101 using transfer learning.
Researched an assistive electric wheelchair that tracks the user’s movements to support people with disabilities. That semester in Japan is where the katakana on this site comes from.
Dean’s List Honoree, Governor’s Award recipient and merit scholarship holder.
Voice-to-voice citizen grievance registration over live phone calls, with structured complaint data extracted into Redis and PostgreSQL.
A RAG assistant over 10 books with book-scoped retrieval, cosine-similarity fallback and source panels for grounded answers.
Real-time intruder detection on webcam and CCTV (RTSP) feeds, with visual and audio alerts below a 0.7 confidence threshold.
More code on GitHub.
I’m Mayank, an AI engineer in Gurugram building production GenAI systems at EaseMyTrip. I mostly work on voice agents that hold real phone conversations, chatbots that call real APIs, and agentic pipelines that do real work. That means caring about latency budgets, hallucination guards and what it costs to serve each call, not just the prompt. My tools are listed on the stack page.
I studied CS with an AI & ML specialisation at NorthCap University and spent a semester in Tokyo, which is why there’s katakana on this site. Outside work you’ll find me in the swimming pool. I’m always happy to talk about voice AI, agents or anything LLM, so feel free to drop me a line.
Send me a message