Deepak Rathore

AI/ML Engineer with 3+ years building and deploying production AI, from embedded firmware and NVIDIA Jetson/TensorRT edge inference in defence R&D (DRDO) to agentic and voice AI systems.

Recent work: a local-first voice AI agent with MCP tool use (LiveKit, Ollama, Whisper), a 10-agent LangGraph pipeline, and a multi-agent ML lifecycle automator for edge robotics. Co-authored 2 research papers; 2nd place in the DRDO-wide Young Scientist AI/ML Challenge. Available immediately for full-time GenAI, Agentic AI and Voice AI roles.

Independent AI Research & Project Development at Independent AI Research & Project Development (2025-10 – Present)

Built 4 production-grade AI systems during self-directed research: Aria (voice AI agent, <1.5s E2E latency), 10-agent LangGraph newsroom pipeline, multimodal RAG system, and 9-agent ML lifecycle automator achieving mAP50=0.804 on NVIDIA Jetson.

Research Fellowship at Aeronautical Development Agency, DRDO (2024-10 – 2025-10)

• Won 2nd place in DRDO-wide Young Scientist AI/ML Challenge — competed against senior scientists and engineers across the entire organisation, solving a physics-based problem using AI/ML
• Integrated Object Detection and Eye Gaze Estimation models into legacy codebase of an active IAF flight simulator system used by pilots and defence personnel
• Developed system automation scripts (Autoboot, Scheduled Run, System Health Check) ensuring 24/7 uptime of mission-critical simulation infrastructure
• Led integration of modern AI/ML workflows into legacy defence systems across the organisation

Started as Trainee Engineer → promoted to full-time Software Engineer → elevated to Research Fellowship at ADA at DEBEL/DRDO (2022-08 – 2024-10)

• Designed and deployed end-to-end AI pipeline for a defence-grade wearable robotics system — spanning sensor selection, embedded firmware, dataset creation, model training, evaluation, and real-time field deployment
• Achieved 98% model accuracy in final real-time field testing across multiple deep learning models
• Developed embedded firmware for IMU sensor data acquisition — high-frequency, low-latency inputs for real-time robotic control
• Implemented real-time inference using Python threading and multiprocessing — separated the Data Acquisition and model inference threads to eliminate blocking, cutting pipeline wait time from 300 ms to 130 ms.
• Deployed deep learning models on NVIDIA Jetson edge devices (TX2, Orin, Nano, Xavier NX) with CUDA-based inference optimization for on-device real-time performance
• Built and processed in-house time-series dataset — full pipeline of cleaning, analysis, visualization, and report generation
• Co-authored 2 research papers on wearable AI; presented monthly to senior scientists, domain engineers (EC/ME/CS/BM), and international teams working on competing technology
• Mentored 5+ junior engineers and interns in model fine-tuning, training, and evaluation

TensorFlow Developer Intern at Tann Mann Foundations (2021-10 – 2022-02)

• Fine-tuned YOLOv5 object detection model achieving 95% efficacy for real-time copyright infringement detection
• Integrated working YOLO v5 model into legacy codebase via Roboflow framework
• Built anti-plagiarism system using computer vision to detect duplicate/copyrighted visual content

Data Analytics Intern at GRRAS solutions Pvt Ltd (2019-06 – 2019-07)

• Implemented end-to-end data engineering pipeline and analytical dashboard using Python and OLAP servers

Bachelors Of Technology in Computer Science & Engineering – Rajasthan Technical University (2017-08 – 2021-10)