Ayura AI
AI-powered health & wellness platform
Personalized Gym, Yoga, Diet, Panchakarma, and Home-Remedy plans via a 4-tier pipeline: rule engine → RAG retrieval → LLM reasoning → safety supervisor.
Case studyOpen to AI / ML roles & internshipsBuilding intelligent systems
ScrollBuilding intelligent systems
with Vision AI & LLMs
Hi, I'm Bhavya Verdia — an AI & Analytics Engineer. I design and deploy production-grade Generative AI agents, RAG pipelines, and computer-vision systems — turning research-grade models into reliable products.
About
Turning models into reliable products.
I'm a final-year Computer Science engineer specializing in AI & Analytics. I love the messy middle between a promising model and a dependable system — evaluation, guardrails, latency, and the last-mile engineering that makes AI trustworthy.
Most recently I interned with Mahindra & Mahindra's Manufacturing IT team, building Vision-AI inspection and LLM agents for the factory floor — hitting 97% detection accuracy in production.
LLM agents & RAG
LangGraph / LangChain multi-agent systems with retrieval and safety gates.
Vision AI
YOLO + OpenCV pipelines for real-world quality inspection.
Production ML
From notebooks to FastAPI services that actually ship.
bhavya@portfolio ~ %
>name:"Bhavya Verdia"
>role:"AI & Analytics Engineer"
>location:"Pune, India"
>education:"MIT ADT University · 2022—26"
>focus:["LLM agents", "RAG", "Vision AI"]
>status:"open to AI/ML roles"
>
97%Vision AI detection accuracy
4+LLM agents designed & shipped
2Full-stack AI platforms built
2026B.Tech CSE — AI & Analytics
Career
Experience & education
Where I've been building, and what I'm studying.
AI Intern, Manufacturing IT
Mahindra & Mahindra Ltd.
- Contributed to a Smart Sequencing Digital Twin for optimizing production flow, including an LLM-based buffer-monitoring agent (LangGraph + XGBoost) forecasting buffer starvation and overflow across TCF shop blocks with real-time PPC alerts.
- Developed a Vision AI system using YOLO to detect wheel center-cap mismatches across multiple Thar SUV variants.
- Implemented colour validation with OpenCV (LAB colour space) plus a transformer-based LLM decision layer (Qwen AI) to handle edge cases and improve pass/fail accuracy.
- Automated the quality-inspection workflow, achieving 97% detection accuracy and reducing manual validation effort.
B.Tech, Computer Science (AI & Analytics)
CurrentMIT ADT University
- Final-year student focused on production AI: LLM agents, RAG pipelines, and Vision AI.
- Strong foundations in DSA, OOP, DBMS, machine learning, and system design.
Selected work
Featured projects
A couple of end-to-end AI systems — tap a card for the deep dive.
More projects
More on GitHub.
Toolkit
Technical skills
The stack I reach for when taking an idea from prototype to production.
AI & Machine Learning
PyTorchScikit-learnYOLOOpenCVNLPGenerative AI
LLM & Orchestration
LangChainLangGraphRAGAzure AI FoundryOpenAI APIs
Backend & Data
PythonJavaFastAPIFlaskREST APIsMySQLMongoDBChromaDB
Tools & Fundamentals
GitDockerPandasNumPyHuggingFace TransformersDSAOOPDBMS
Contact
Let's build something great.
I'm open to AI/ML roles, internships, and collaborations. Drop me a message and I'll get back to you soon.