AI/ML Engineer · LLM Systems · RAG · Voice & Agentic Pipelines
Building things that actually work — then writing about them so others can too.
I build LLM-powered systems end-to-end — from fine-tuning large models on custom datasets to deploying them in voice bots, WhatsApp assistants, and agentic pipelines. I care about making AI work in production, not just in notebooks.
Right now I'm at Oriserve as a Data Scientist, working on:
- 🎙️ Speech-to-speech streaming pipelines with multimodal LLMs
- 🔧 Fine-tuning Mistral 24B on multi-GPU AWS H100s (saved ~$92K/yr in API costs)
- 📦 Vector DB management, LLM migration, evaluation, & system optimization for enterprise chatbots
Previously at Namasys Analytics, I built AI Ops evaluation pipelines, agentic voicebots with AI avatars, and GPU-accelerated supply chain optimizers for Middle-Eastern clients.
LLMs & AI
OpenAI Mistral Gemini LangChain PEFT RAG Fine-tuning Agentic Pipelines
Voice & Multimodal
Voxtral Whisper TTS VAD Rnnoise Anam
Stack
Python FastAPI React Next.js ChromaDB Pinecone Weaviate Docker AWS
| Project | What it is | Stack |
|---|---|---|
| rag-evaluation | RAG pipeline with SSE streaming + eval dashboard | ChromaDB · OpenAI · FastAPI · React |
| rag-vs-agents-tabular | RAG vs Agents on tabular/CSV data — side by side | Pandas · LangChain · FastAPI |
| Layla (Dubai Leasing AI) | Voice + WhatsApp lead assistant with conversation state machine | FastAPI · Twilio · GPT-4o · RAG |
| Parkify | Parking app with custom Indian number plate OCR | OpenCV · TensorFlow · Flask |
🏆 2nd place @ Hack4Bengal 2.0 · 🏆 2nd place @ Diversion 2k23 (MLH)
I post regularly on LinkedIn about AI/ML projects — practical stuff, real code, no fluff. If you're building with LLMs, we probably have things to talk about.
Open to interesting problems, collabs, or just a good AI conversation. aryachakraborty2002@gmail.com