AI/ML Engineer based in Bangalore, India — building production-grade LLM systems, voice AI pipelines, and real-time backends.
I work across the full AI stack: from fine-tuning open-source LLMs and building multi-turn voice booking engines to shipping real-time tracking systems used by live users. My work lives in production — not in notebooks.
Current focus: LLM pipelines · Voice AI · AWS Bedrock · LangGraph · RAG systems
| Repo | What it does | Stack |
|---|---|---|
| 🤖 qwen-fine-tuning-pipeline | Fine-tuned Qwen 2.5 for intent classification across 5 service categories — 13K labeled rows, PEFT/LoRA, full validation suite | Python · Hugging Face · PEFT · Pandas · Qwen 2.5 |
| 🎙️ llm-voice-booking-engine | Multi-turn voice booking engine with 9-stage LLM pipeline and LangGraph state machines on AWS Bedrock | Python · LangGraph · AWS Bedrock · WebSocket · FastAPI |
| 🛡️ crisis-detection-classifier | Two-layer crisis detection (regex + LLM fallback) with safety gate and emotion/tone engine — 145+ tests passing | Python · PyTorch · AWS Bedrock · FastAPI · pytest |
AI/ML AWS Bedrock (Claude) · LangGraph · Qwen 2.5 fine-tuning · Hugging Face PEFT
RAG pipelines · Intent Classification · Emotion Detection · NLP
PyTorch · TensorFlow · Whisper STT · TTS · VAD · ElevenLabs
Voice AI LiveKit · SIP Telephony · librosa · torchaudio · ElevenLabs
Backend Node.js · TypeScript · FastAPI · Flask · Socket.io · Redis · REST APIs
WebSocket · Zod · Async Programming · PM2
Frontend React Native · Expo · Context API · Expo Push Notifications
Cloud/DevOps AWS (Bedrock · RDS · CloudWatch · S3) · Azure · Docker · GitHub Actions
Associate AI/ML Engineer @ Yugasys Software (HeiBuddy AI) · Feb 2026 – Present
- Shipped multi-turn LLM voice booking engine on AWS Bedrock with 9-stage processing pipeline
- Built 4-phase Recommendation Engine with user affinity scoring, geo-ranking, and Redis caching
- Implemented two-layer Crisis Detection + Emotion & Tone Engine with 145+ passing tests
- Built Qwen 2.5 fine-tuning pipeline — 13K labeled rows, Excel-to-JSONL converter, validation suite
Associate Software Developer – AI/ML @ TVAM Technologies · Jan 2025 – Jan 2026
- Built real-time voice AI pipelines: Whisper STT + TTS + VAD + SIP telephony for 1,000+ merchants
- Designed RAG retrieval system with Redis caching — reduced API latency by ~40%
- Deployed LiveKit for real-time audio on Azure; containerized with Docker + GitHub Actions
