profile = {
    "name": "Vivek Sawant",
    "roles": [
        "Applied Scientist (ex-Amazon)",
        "Industrial Design Engineer",
        "AI Systems Builder"
    ],
    "focus": {
        "research":     ["CNNs", "Vision Transformers", "RAG Systems", "CRAG"],
        "engineering":  ["Computer Vision Pipelines", "Edge AI", "Robotics Perception"],
        "design":       ["Human-Centered Systems", "Product Architecture", "UX-Aware ML"],
    },
    "current_mission": "Designing AI systems that are not just accurate — but deployable, usable, and impactful.",
    "principle":       "Every model is a product. Every product is a system.",
}

• Amazon Applied Scientist Intern — Worked on ML systems in an applied research environment at one of the world's most data-intensive organizations
• Samsung Solve for Tomorrow — National Finalist; designed a technology-driven solution evaluated among India's top student innovations
• Flipkart Grid 7.0 — Semifinalist in India's largest engineering challenge for students, competing across AI/robotics tracks
• Amazon ML Summer School — Selected for Amazon's competitive ML program covering advanced topics in deep learning and applied science
• HPAIR Tokyo — Delegate at Harvard's flagship Asia conference; engaged with global leaders across technology and policy

AI / Machine Learning PyTorch TensorFlow Transformers (HuggingFace) CNNs Vision Transformers RAG CRAG Scikit-learn ONNX

Computer Vision & Design OpenCV YOLO Detectron2 Roboflow Figma SolidWorks Fusion 360 Blender

Data / Backend / Tools Python FastAPI LangChain FAISS PostgreSQL Docker Git Linux Jupyter W&B

Autonomous patient intake and preliminary diagnostic system using computer vision and NLP pipelines.

Architected an end-to-end kiosk that combines face-based vitals estimation, symptom parsing, and triage classification — designed for deployment in low-resource healthcare settings. Stack: OpenCV · PyTorch · FastAPI · Raspberry Pi

Robotics-integrated dispensing system with vision-guided object recognition and actuation control.

Built a perception-to-action pipeline enabling real-time product identification and precise dispensing — bridging embedded systems with trained CV models. Stack: YOLO · Arduino/ROS · Python · Custom PCB Design

Corrective Retrieval-Augmented Generation system for high-fidelity, hallucination-resistant QA.

Implemented a CRAG pipeline with dynamic retrieval grading and query refinement, significantly improving answer faithfulness over naive RAG baselines. Stack: LangChain · FAISS · HuggingFace · GPT-4 · Python