End-to-End Production Experience: I don’t just build models in notebooks; I ship them. I have extensive experience building and deploying machine learning and computer vision models as scalable FastAPI endpoints using Docker and AWS (Lambda, EC2, ECR).
Cutting-Edge Generative AI & VLMs: I specialize in fine-tuning and quantizing Vision-Language Models (VLMs) and LLMs for domain-specific tasks. For a recent project, I improved extraction accuracy by 30% using advanced prompt engineering and optimized workflows.
Specialized Document Intelligence: I can build complex “Document Intelligence” systems. My work includes developing multilingual pipelines that use OCR and GenAI to extract and validate data from intricate transaction records like invoices and shipping documents.
Real-World Computer Vision Applications: From real-time video analytics for retail (person tracking and behavior analysis) to automated car damage detection using YOLOv8 and semantic segmentation, I build CV systems that solve high-stakes physical world problems.
Automated Retraining & Data Pipelines: I focus on long-term model health by building robust data pipelines and implementing automated retraining and classification pipelines to handle data drift and skewed distributions.
Strategic AI Leadership: Beyond coding, I lead AI-ML teams and collaborate directly with product and business stakeholders. This means I understand how to translate your business requirements into technical KPIs that actually improve your bottom line.
Experience Level
Language
Work Experience
Education
Qualifications
Industry Experience
Developed an automated car damage detection system using OpenCV, semantic segmentation, and YOLOv8 models, deployed
via FastAPI endpoints; optimized inference speed by 30%, enabling real-time panel-level analysis from morning and evening
vehicle images.
Implemented semantic segmentation techniques alongside object detection to localize and delineate damaged regions with
greater precision, improving detection accuracy by 25% under varying lighting and occlusion conditions.
Integrated LLMs and GenAI models (e.g., Gemini 1.5 Flash) to assess damage severity, classify affected panels, and generate
contextual insights, reducing manual inspection efforts by 40%.
Designed and deployed a comprehensive reporting pipeline for service personnel, summarizing damage type, intensity, and
recommended actions; end-to-end workflow efficiency improved by 35%.
Developed a real-time video analytics system using RTSP stream port-forwarded to DigitalOcean, leveraging NVIDIA DeepStream
on Jetson Orin for person detection, tracking, age & gender classification, and behavior analysis.
Implemented region-based dwell time, footfall, entry-exit detection, and demographic insights using NVIDIA’s pre-trained models,
and YoloV8 finetuned models, enabling accurate in-store movement analysis.
Aggregated detection metrics on hourly and daily intervals to generate actionable KPIs for retail dashboards, enhancing visibility
into customer flow and engagement zones.
Integrated real-time alerting system via Telegram and email based on dwell time thresholds, enabling store managers to respond
instantly to crowding or long wait times.
Working on VeraAI, a multilingual document intelligence system leveraging advanced OCR engines and Vision-Language Models
(VLMs) for robust text extraction and translation across complex document formats.
Improved extraction and translation accuracy by 30% over traditional methods through advanced prompt engineering and
optimized model workflows.
Currently fine-tuning and quantizing VLMs for efficient inference, with deployment on AWS Lambda, Sagemaker and EC2 to
ensure scalable, low-latency processing.
Supporting 13+ languages and multiple file formats (PDF, images, Excel, Word), while developing domain-specific validation
algorithms for accurate field-level verification in transaction documents.
Developed a computer vision based document classification system using ML models to automatically categorize uploaded
documents (Invoice, Shipment, PO, TC, etc.), reducing dependency on manual verification.
Built and deployed the classification service as scalable APIs using FastAPI, containerized with Docker, and integrated with AWS
Lambda for event-driven, serverless processing.
Implemented class balancing and optimized model training pipelines to handle skewed document distributions, improving
classification accuracy and minimizing misclassification across document types.
Integrated an automated alerting mechanism to notify suppliers for incorrect document uploads, significantly reducing manual
effort for the customer engagement team and improving overall workflow efficiency.
Developed the UTrust Price Recommendation–Lead Score App, using regression algorithms to predict optimal second-hand car
prices and classification models to estimate customer engagement likelihood; improved price prediction accuracy by 28% and
lead classification precision by 32%.
Built a customer audio analytics (SBERT- RoBERTa), feature engineering,
normalization, and encoding, enabling high-quality model inputs and reducing preprocessing time by 35%.
Trained and fine-tuned multiple regression models (Linear Regression, Decision Tree Regressor, MLP) and classification models
(Logistic Regression, SVM, Decision Tree Classifier), selecting best-performing models through hyperparameter tuning and cross
validation.
Delivered an intuitive app interface for procurement officers and sales teams, offering instant, data-driven price evaluations and a
lead scoring system that improved sales targeting efficiency by 40%, streamlining procurement and conversion workflows.
Hire a Data Scientist
We have the best data scientist experts on Twine. Hire a data scientist in Bengaluru today.