I am a Machine Learning Engineer focused on bridging the gap between research models and real-world applications. With a strong foundation in Azure AI and Statistical Modeling, I build scalable AI systems—from training custom LLMs to deploying them on edge devices or modern web architectures.
- 🔭 Current Focus: GraphRAG, Local LLM Agents, Privacy-first Edge AI
- 💼 Experience: Former Data Scientist Intern at iNeuron.ai and ML Intern at F(x) Data Labs
- ☁️ Cloud Native: Model deployment using Docker and Azure App Services
Azure AI Engineer Associate (AI-102)
Skills: NLP, Computer Vision, Generative AI
Azure Data Scientist Associate (DP-100)
Skills: Model Training, MLOps, Data Pipelines
My core toolkit for extracting insights and building intelligence.
| Core AI/ML | Data Ops & Cloud | Deployment (The Engineering Layer) |
|---|---|---|
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
Focus: Real-Time Orchestration, Event-Driven Architecture, Local RAG
A production-grade voice orchestrator designed to automate complex hospitality workflows, bridging generative conversation with deterministic business logic.
- <500ms Latency: Engineered a high-throughput voice pipeline using EdgeTTS and Gemini 2.0 Flash Lite for near-instant responses.
- Hybrid Intelligence: Dynamically routes user intent between Tool Calling (Booking System), Vector Search (ChromaDB for Policy RAG), and General Chat.
- Event-Driven CRM: Implemented an asynchronous logging pipeline that syncs structured call summaries to Google Sheets without blocking the UI thread.
- Tech: Next.js, FastAPI, Gemini 2.5 Flash, ChromaDB, EdgeTTS, WebSockets
🔗 https://github.com/DevanshMistry890/hotel-voice-agent
Focus: Edge Computing, Privacy, Transformers.js
A privacy-first sentiment analysis tool that runs entirely in the browser.
- Runs quantized ONNX models locally
- Zero backend, zero latency, zero data leakage
- Tech: Transformers.js, React, Web Workers
🔗 https://github.com/DevanshMistry890/edge-nlp-analyzer
Focus: Generative AI, Vector Search, RAG
A Retrieval-Augmented Generation system that creates personalized recipe recommendations.
- Reduces hallucinations using vector-grounded responses
- Tech: Google Gemini API, MongoDB Atlas Vector Search, Next.js
🔗 https://github.com/DevanshMistry890/recipebook
