Priyanshu Kumar
+91-781******* *******************@*****.*** linkedin.com/in/priyanshukumar062003 github.com/Priyanshukumarx Technical Skills
Computer Vision: OpenCV, YOLOv8, Detectron2, InsightFace, Object Detection, Ball Tracking, Player Tracking, Pose Estimation, Re-Identification (ReID)
Deep Learning & ML: PyTorch, TensorFlow, CNN, LSTM, TCN, CatBoost, Feature Engineering, Data Augmentation, Model Evaluation
Tracking & Video Analytics: Multi-Object Tracking, Trajectory Analysis, Motion Analysis, Bounce Detection, Temporal Event Detection, Homography, Perspective Transformation
Model Optimization & Deployment: ONNX, TensorRT, CUDA, FP16, ONNX Runtime, GPU-Accelerated Inference, GStreamer
Edge AI & On-Device Processing: NVIDIA Jetson Orin, Edge Inference, Multi-Camera Video Processing, Real-Time Video Processing
Generative AI: Diffusion Models, DDPM, CLIP, Video U-Net, LLM Quantization Vector Search & Embeddings: FAISS, Embedding Similarity, Vector Retrieval Data Processing: NumPy, Pandas, Scikit-learn
Backend & APIs: FastAPI, Node.js, Express.js, REST APIs Frontend & Languages: Python, C++, C#, JavaScript, TypeScript, React, React Native, Redux Toolkit Work Experience
Vareli Tecnac Pvt. Ltd. Kolkata, India
AI/ML Engineer Sep 2025 – Present
– Optimized inference pipelines, increasing throughput from 7 FPS to 60 FPS (8.6 improvement) and cutting processing time for a 30-second video from 5 minutes to 1 minute (80% reduction).
– Developed real-time computer vision pipelines for object detection, ball tracking, player tracking, and pose estimation using deep-learning models.
– Developed and optimized model architecture for NVIDIA Jetson Orin, enabling real-time, on-device processing of multi-camera video streams under edge-computing constraints.
– Optimized model deployment using PyTorch, ONNX, TensorRT, CUDA, and FP16 acceleration for low-latency inference.
– Implemented trajectory analysis, motion analysis, and event detection (including LSTM/TCN-based ball bounce detection) using frame-wise coordinates, homography and perspective transformation, feature engineering, and machine-learning classification.
– Developed multi-camera face-recognition and identity-matching pipelines using InsightFace, FAISS, embedding similarity, and GStreamer; worked on multi-object tracking and player re-identification across frames and camera views.
– Developed embedding similarity and VIP intelligence workflows for retail analytics using high-performance vector retrieval, and optimized GPU-accelerated deployment components for stable, long-duration video processing.
– Built modular LSTM/CatBoost training pipelines (preprocessing, augmentation, validation) and a background-running C#/.NET asset-tracking app using REST APIs and SMBIOS UUID/SHA-256 machine identification. Research & Personal Projects
Text-to-Video Diffusion Generator PyTorch, Video U-Net, DDPM Jan 2026 – Present
– Built a lightweight text-conditioned video diffusion pipeline using PyTorch.
– Implemented CLIP-based text embeddings and a custom Video U-Net; integrated DDPM reverse diffusion with classifier-free guidance.
7B Diffusion LLM Exploration Dream 7B, LLM, Quantization Dec 2025
– Explored diffusion-based language modeling with Dream 7B and applied 4-bit quantization for low-VRAM GPU experimentation.
VCam Studio - Virtual Camera Simulator OpenCV, PyQt6 Nov 2025
– Developed an interactive 6-DOF virtual camera system with intrinsic & extrinsic parameter control, lens distortion modeling, and perspective warping.
Certifications
OCI 2025 Generative AI Professional Certification Oracle Aug 2025 Build AI Apps with ChatGPT, Dall-E, and GPT-4 (ID: 3DWD7EQVND2B) Coursera 2024 Education
Lovely Professional University Phagwara, Punjab
Bachelor of Technology, Computer Science and Engineering Aug 2022 – May 2026