Open to opportunities

Deep Mathukiya

@deepmathukiya

Message

I build deployable AI agents, computer vision systems, and forecasting tools with measurable performance.

India

Message

What I'm looking for

I'm looking to build and deploy practical AI systems across LLM agents, computer vision, and forecasting, where I can turn research into measurable real-world outcomes.

At HyperNorm AI, I architected a controlled, low-cost Web Search MCP tool for LLM integration and engineered an observable LLM gateway with RBAC, rate and token limiting, and roughly 10 ms added latency. I also optimized a LangGraph financial research agent with GEPA, outperforming Perplexity across 120 queries.

I've built production-minded AI systems across computer vision, language, and forecasting: a Gujarati Sign Language-to-Speech translator with 3,600+ gestures and 94% accuracy, a YOLO and InsightFace recognition pipeline, and ARIMA/LSTM temperature forecasting using 45+ years of meteorological data. My work has earned an NVIDIA Nemotron 3 Super Contest Hackathon win, a PDEU IIC grant, and a patent for an offline sign language and speech translator.

Experience

Work history, roles, and key accomplishments

Current

ML Intern

Current

HyperNorm AI

Dec 2025 - Present (8 months)

Architected a fully controlled, low-cost Web Search MCP tool for seamless LLM integration. Engineered LLM gateway with RBAC, 100% observability, rate/token limiting, and adding ~10 ms latency.

MCP LLM RBAC Observability Rate Limits LangGraph GEPA

AI Trainee

Buze Platforms Private Limited

May 2025 - Jul 2025 (2 months)

Built YOLO + InsightFace person recognition pipeline with 92% accuracy at 10 FPS. Developed low-latency C++ face recognition on Rockchip RV1126 using RKNN and MPP with IPC.

YOLO InsightFace C++MPP IPC NVIDIA T4

Research Intern

India Meteorological Department

May 2025 - Jul 2025 (2 months)

Developed time series models (ARIMA, LSTM) for long-term temperature forecasting with RMSE of 1.18°C. Processed over 45 years of meteorological data for model training.

ARIMA Time Series Modeling Data Processing

Student Researcher

Pandit Deendayal Energy University

Oct 2023 - May 2025 (1 year 7 months)

Led SignToSpeech, a Gujarati Sign Language to Speech translator using 3,600+ gestures with 94% accuracy. Deployed on NVIDIA Orin Nano with under 500 ms latency and awarded a ₹62K grant and patent.

Deep Learning Computer Vision NVIDIA Orin Sign Language Recognition