I’m an AI Engineer who builds production-grade generative AI and retrieval systems — not prototypes with no impact.

I specialize in:

• Architecting and operating production RAG pipelines for real-time semantic search
• Designing LLM routing and grounding systems with reliability and observability
• Building scalable backend APIs for AI workflows using FastAPI and async patterns
• Evaluating LLMs for cost, latency, and consistency

🔭 Current Focus: Production RAG, LLM routing & query classification, LLM evaluation metrics, multi-agent orchestration

🚀 Target Roles: AI Engineer · Generative AI Engineer · LLM / RAG Systems Engineer · Applied Machine Learning Engineer

Python · FastAPI · Async APIs · REST · Server-Sent Events (SSE) · Docker · SQL · PostgreSQL

Generative AI · Retrieval-Augmented Generation (RAG) · Semantic Search · Large Language Models (LLMs) · LLM Evaluation & Trustworthiness Scoring LangChain · LangGraph · MCP · Prompt Engineering · Embeddings & Vector Search

ETL Pipelines · ChromaDB · FAISS · Supabase · AWS (EC2, S3) · CI/CD

Streamlit · Dashboards · SHAP Interpretability

GitHub

Production-grade retrieval system combining hybrid sparse + dense search across text and images, with grounded citations and real-time streaming APIs.

• Hybrid retrieval: BM25 + vector embeddings for multi-modal semantic relevance
• Grounded citation outputs for traceable answers
• FastAPI based streaming interface using SSE
• LLM benchmarking across cost, latency, and accuracy
• Async workers, caching, and scalable backend design

What I Learned

• Hybrid retrieval architecture
• Multi-modal semantic search
• Building real-time streaming APIs
• Structuring benchmarking pipelines for LLM evaluation

GitHub

Framework to assess LLM trustworthiness and behavioral consistency without ground-truth labels.

• Reference-free scoring for LLM outputs
• Behavioral consistency and multi-choice evaluation logic
• Modular Python evaluation pipelines
• Designed for reliability analysis in production-like setups

What I Learned

• LLM evaluation workflows
• Reference-free reliability metrics
• Extensible Python pipelines
• Challenges in model trustworthiness

GitHub

LangGraph-based CLI client coordinating multiple MCP servers for action routing and multi-tool workflows.

• Tool orchestration (notes, weather, web search, automation)
• Session state management and contextual routing
• Extensible plugin-style MCP server integration

What I Learned

• Multi-agent orchestration patterns
• Protocol integration with MCP
• Session tracking and routing logic
• Structuring agent toolchains

Streamlit App | GitHub

Unsupervised patient stratification with HDBSCAN and SHAP-powered interpretability dashboards.

• 130k+ patient record clustering
• SHAP interpretation and interactive exploration
• End-to-end pipeline from data to UI

IEEE Paper

A multimodal system detecting beehive health via vision and audio, published at IEEE CSCITA 2023.

• Integrated YOLO vision (96.2% precision) and audio classification (99.8%)
• Real-world deployment mindset
• Hackathon winning submission

• 🥇 1st Place — Voxel51 Visual AI Hackathon | GitHub
• 📜 AWS Certified Cloud Practitioner | Credly Badge

• 🔗 LinkedIn: https://linkedin.com/in/deepmehta27
• 🖥️ GitHub: https://github.com/deepmehta27