Oct 2023 – Aug 2025 · 1 yr 10 mos · Noida, Uttar Pradesh, India · Hybrid

• As the sole ML engineer at IntelliRail, I built next-gen AI systems combining Generative AI and Computer Vision to transform railway and warehouse safety. My work fused edge AI pipelines, LangGraph multi-agent orchestration, and cloud analytics to deliver measurable operational impact.
• 📊 UltraTech Cement – Agentic Work-Order Automation: Built a LangGraph-powered multi-agent workflow that transformed incidents, SOPs, and inventory data into schema-validated work orders with ETAs, checklists, and Jira/ServiceNow tickets. The pipeline integrated Postgres incidents, GCS evidence, SOP/Manual PDFs, and parts/shift calendars, enforcing schema validation and human-in-loop approvals. Impact: cut work-order creation time by ~80% (minutes → seconds) and reduced repeat incidents by 20–30% through similar-case recommendations and SOP citations.
• 📦 Amazon Warehouses – Asset Monitoring: Designed a multi-camera DeepStream + Jetson system for tray/personal-belongings validation at exits. Integrated YOLO detection, PGIE/SGIE classification, OCR for usernames, and retry-enabled QR code reading with employee-ID validation. Added a LangGraph-powered “Explain with AI” copilot for one-click incident reports, shift summaries, and repeat-offender alerts. Results: 80% fewer losses, 25% higher belonging recovery, >90% detection accuracy.
• 🖥 Edge AI & CV Pipelines: Optimized DeepStream/TensorRT/YOLO pipelines on Jetson AGX for millisecond-level inference across multiple streams.
• 🔍 Automated Defect Detection: Built CV algorithms for bulge detection in rolling stock, enabling predictive maintenance.
• 📷 Cloud & Multi-Camera Orchestration: Integrated camera networks with GCP (Postgres + GCS + pgvector) for evidence storage, retrieval, and analytics.
• ⚡ Operational Integration: Linked inference + agent outputs with REST APIs, dashboards, and ticketing—turning manual, reactive checks into proactive, AI-powered intelligence platforms.

Jan 2022 – Sep 2023 · 1 yr 8 mos · Noida, Uttar Pradesh, India · Remote

• As a Machine Learning Engineer, I designed and deployed ML and Computer Vision solutions that digitized inspection workflows, automated repetitive tasks, and improved operational efficiency. My work spanned end-to-end pipelines—from data acquisition and model training to embedded deployments and integration with enterprise systems.
• 📄 OCR & Information Extraction – Built TfLite-based OCR pipelines (96% accuracy, 70% fewer recognition errors) to extract structured information from IDs and asset labels, improving data reliability and reducing manual entry.
• 🎯 Object Detection & Classification – Developed TensorFlow models for detecting and classifying key components in complex environments, enabling proactive monitoring, faster inspection cycles, and predictive maintenance planning.
• 🖥 Edge AI Deployment – Optimized and packaged models for Raspberry Pi and NVIDIA Jetson devices, delivering low-latency, on-device inference suitable for field operations in remote or bandwidth-limited settings.
• 🔌 Automation & Data Pipelines – Streamlined data ingestion, preprocessing, and reporting workflows with Python, TensorFlow, and cloud services, reducing response times by 83% and enabling real-time insights for operators.
• 📊 Enterprise Integration – Connected ML inference results with internal databases, dashboards, and reporting tools, ensuring seamless adoption into existing business processes.
• Through these projects, I contributed to the transition from manual inspections to automated, AI-powered monitoring systems, establishing a foundation for data-driven decision-making and large-scale operational improvements.

Jun 2021 – Nov 2021 · 5 mos · Remote

• At Vybo.io, I contributed to document digitization and AI-powered content cleaning, developing deep learning solutions for handwritten document processing as part of the company’s Document Cleaning Module.
• 📝 Rules Removal in Handwritten Documents: Developed a deep learning-based computer vision model to remove ruled lines in handwritten documents, enhancing readability for downstream OCR systems. Achieved 95%+ accuracy on targeted cases.
• 🧪 Synthetic Data Generation: Created large-scale synthetic datasets of handwritten documents to improve model robustness, reducing dependency on scarce real-world training data.
• 🔄 Iterative Model Optimization: Performed multiple training–testing cycles, fine-tuning parameters and architectures to maximize accuracy and minimize false removals.
• 👥 Annotation Pipeline Management: Coordinated an iterative data annotation process with input from 100+ freelancers and annotators, ensuring high-quality training data.
• 🔍 Model Testing & Error Analysis: Tested trained models on scenario-specific datasets, identified edge cases, and proposed targeted improvements to boost performance.
• This role strengthened my expertise in document image processing, data pipeline orchestration, and annotation workflow management, laying the foundation for my later computer vision projects.