Arkadeep D.

Searching for a 𝐌𝐚𝐜𝐡𝐢𝐧𝐞 𝐋𝐞𝐚𝐫𝐧𝐢𝐧𝐠 𝐄𝐧𝐠𝐢𝐧𝐞𝐞𝐫 who doesn't just build models but engineers real-world results? When deadlines are tight, data is messy, and the expectations are sky-high, what you need is precision. You need a 𝐌𝐚𝐜𝐡𝐢𝐧𝐞 𝐋𝐞𝐚𝐫𝐧𝐢𝐧𝐠 𝐄𝐧𝐠𝐢𝐧𝐞𝐞𝐫 who brings applied intelligence to every frame, feature, and inference pipeline, and someone who builds systems that run in production, not just on paper. I deliver end-to-end solutions across 𝐋𝐋𝐌 𝐩𝐢𝐩𝐞𝐥𝐢𝐧𝐞𝐬, 𝐜𝐨𝐦𝐩𝐮𝐭𝐞𝐫 𝐯𝐢𝐬𝐢𝐨𝐧, 𝐚𝐧𝐝 𝐀𝐈 𝐟𝐨𝐫 𝐨𝐩𝐞𝐫𝐚𝐭𝐢𝐨𝐧𝐬, all built to deliver impact. Whether it's GPT-4o-powered automation or YOLO-based safety systems running on edge devices, every project is shaped by problem-first thinking and production-focused execution. 𝐖𝐡𝐚𝐭 𝐝𝐨𝐞𝐬 𝐭𝐡𝐢𝐬 𝐌𝐚𝐜𝐡𝐢𝐧𝐞 𝐋𝐞𝐚𝐫𝐧𝐢𝐧𝐠 𝐄𝐧𝐠𝐢𝐧𝐞𝐞𝐫 𝐃𝐞𝐥𝐢𝐯𝐞𝐫? ▶ 𝐋𝐚𝐧𝐠𝐮𝐚𝐠𝐞 𝐌𝐨𝐝𝐞𝐥𝐬 𝐓𝐡𝐚𝐭 𝐖𝐨𝐫𝐤 𝐋𝐢𝐤𝐞 𝐓𝐞𝐚𝐦𝐦𝐚𝐭𝐞𝐬 Deployed GPT-4o and Llama 3-based multi-agent systems for project planning, cost estimation, and operational alignment, resulting in up to 𝟒𝟔% 𝐠𝐚𝐢𝐧𝐬 𝐢𝐧 𝐞𝐟𝐟𝐢𝐜𝐢𝐞𝐧𝐜𝐲, 𝐚𝐧𝐝 𝐦𝐨𝐫𝐞 𝐭𝐡𝐚𝐧 𝟐𝟕% 𝐬𝐚𝐯𝐢𝐧𝐠𝐬 𝐢𝐧 𝐩𝐫𝐨𝐣𝐞𝐜𝐭 𝐜𝐨𝐬𝐭𝐬. ▶ 𝐕𝐢𝐬𝐢𝐨𝐧 𝐒𝐲𝐬𝐭𝐞𝐦𝐬 𝐟𝐨𝐫 𝐒𝐚𝐟𝐞𝐭𝐲, 𝐒𝐞𝐜𝐮𝐫𝐢𝐭𝐲 & 𝐕𝐞𝐫𝐢𝐟𝐢𝐜𝐚𝐭𝐢𝐨𝐧 From 𝐘𝐎𝐋𝐎𝐯𝟖 𝐨𝐧 𝐉𝐞𝐭𝐬𝐨𝐧 𝐍𝐚𝐧𝐨 𝐭𝐨 𝐏𝐚𝐝𝐝𝐥𝐞𝐎𝐂𝐑 𝐩𝐢𝐩𝐞𝐥𝐢𝐧𝐞𝐬, I have delivered video analytics systems that detect helmets, masks, fire, smoke, car crashes, and national ID cards, deployed in safety-critical and financial environments. ▶ 𝐑𝐀𝐆 & 𝐋𝐋𝐌-𝐃𝐫𝐢𝐯𝐞𝐧 𝐈𝐧𝐭𝐞𝐫𝐟𝐚𝐜𝐞𝐬 𝐓𝐡𝐚𝐭 𝐑𝐞𝐩𝐥𝐚𝐜𝐞 𝐌𝐚𝐧𝐮𝐚𝐥 𝐋𝐚𝐛𝐨𝐫 Built hybrid retrieval + generation models using 𝐯𝐞𝐜𝐭𝐨𝐫 𝐝𝐚𝐭𝐚𝐛𝐚𝐬𝐞𝐬 (𝐌𝐢𝐥𝐯𝐮𝐬, 𝐏𝐢𝐧𝐞𝐜𝐨𝐧𝐞) 𝐚𝐧𝐝 𝐋𝐚𝐧𝐠𝐂𝐡𝐚𝐢𝐧/𝐋𝐚𝐧𝐠𝐆𝐫𝐚𝐩𝐡, automating estimation workflows, reducing human effort by 67%, and accelerating planning cycles by 36%. ▶ 𝐄𝐝𝐠𝐞-𝐑𝐞𝐚𝐝𝐲, 𝐋𝐨𝐰-𝐋𝐚𝐭𝐞𝐧𝐜𝐲 𝐀𝐈 𝐒𝐲𝐬𝐭𝐞𝐦𝐬 Trained and optimized detection models to run in real-time on 𝐉𝐞𝐭𝐬𝐨𝐧 𝐍𝐚𝐧𝐨, delivering fast, localized processing without cloud dependency. ▶ 𝐒𝐞𝐜𝐮𝐫𝐞 𝐀𝐈 𝐏𝐢𝐩𝐞𝐥𝐢𝐧𝐞𝐬 𝐟𝐨𝐫 𝐎𝐧𝐥𝐢𝐧𝐞 𝐁𝐚𝐧𝐤𝐢𝐧𝐠 𝐚𝐧𝐝 𝐅𝐢𝐧𝐭𝐞𝐜𝐡 Developed Siamese and segmentation models to verify ID cards and MRZ codes with over 𝟗𝟔% 𝐚𝐜𝐜𝐮𝐫𝐚𝐜𝐲, improving fraud detection in real-time loan applications. ▶ 𝐎𝐂𝐑 𝐚𝐧𝐝 𝐃𝐨𝐜𝐮𝐦𝐞𝐧𝐭 𝐈𝐧𝐭𝐞𝐥𝐥𝐢𝐠𝐞𝐧𝐜𝐞 𝐟𝐨𝐫 𝐄𝐧𝐠𝐢𝐧𝐞𝐞𝐫𝐢𝐧𝐠 & 𝐋𝐞𝐠𝐚𝐥 𝐖𝐨𝐫𝐤𝐟𝐥𝐨𝐰𝐬 Built custom UNet models and image segmentation logic to extract data from complex layouts like callouts, title blocks, and engineering diagrams, increasing accuracy up to 93.47%. 𝐓𝐞𝐜𝐡𝐧𝐢𝐜𝐚𝐥 𝐒𝐭𝐚𝐜𝐤 ▶ 𝐋𝐚𝐧𝐠𝐮𝐚𝐠𝐞𝐬: Python, C++ ▶ 𝐌𝐋 𝐅𝐫𝐚𝐦𝐞𝐰𝐨𝐫𝐤𝐬: PyTorch, TensorFlow, Scikit-learn ▶ 𝐍𝐋𝐏 & 𝐋𝐋𝐌𝐬: GPT-4o, LLaMA 3, LangChain, LangGraph, RAG pipelines ▶ 𝐂𝐨𝐦𝐩𝐮𝐭𝐞𝐫 𝐕𝐢𝐬𝐢𝐨𝐧: YOLOv8, UNet++, Siamese Networks, OpenCV ▶ 𝐎𝐂𝐑 𝐓𝐨𝐨𝐥𝐬: PaddleOCR, Tesseract, Amazon Textract ▶ 𝐀𝐈 𝐎𝐩𝐬: FastAPI, Docker, Git ▶ 𝐕𝐞𝐜𝐭𝐨𝐫 𝐃𝐁𝐬: Milvus, Pinecone ▶ 𝐀𝐩𝐩𝐥𝐢𝐜𝐚𝐭𝐢𝐨𝐧 𝐀𝐫𝐞𝐚𝐬: LLMs, Computer Vision, OCR, AI on Edge, Fintech Security 𝐑𝐞𝐜𝐨𝐠𝐧𝐢𝐭𝐢𝐨𝐧𝐬 ▶ Consistently delivered solutions that reduced manual effort by over 60% ▶ Achieved production accuracy benchmarks of 𝟗𝟑–𝟗𝟔% across AI projects deployed in live systems