Nitin Sharanarthi Email & Phone Number

Bengaluru, Ka, In

India

1. OCR on sticker and surface level defect detection on metal surface using Yolov52. Oil cooler and Gear inspection using Yolv8 Technologies used to build end-to-end platform includes : Django backend framework, Kafka(streaming), Redis/ArticDb(cache), MongoDb(primaryDb), Nginx load balancer, Celery, FastAPI, Docker, GCP cloudR&D projects : 1. PCB AOI : Detecting defects on PCB surface on both primary and secondary sideAcquisition device used : 3d scanners wenglor / lmi : Algorithms/lib on 3d point cloud data : DBSCAN, RANSACAnomalib STFPN for anomaly region highlight faulty regions.

India

Deployed a complete packaged AI solution to a leading electronic component manufacturer.• Designed the pipeline/flow, wrote backend and linked APIs with SOTA detection models available at that time (TensorFlow Object Detection API) Model used: FRCNN with inceptionv4 backbone.• The entire end-to-end training platform was built and provided to the customer. • He could create workstations, add cameras to it, collect data, annotate, and train several versions and choose to deploy the one which has the highest accuracy.• Custom logic was built according to the customer needs from model output. • Reduced cycle Time by Pruning and running several instances of model to run in parallel.Result: More than 1000+ inspections done per month from past 2021, user can train and deploy models when new parts arrive from a different vendor without our interventionOther Projects/POCs:1. Segmentation model to detect outflow/inflow of liquid pour : model used DeepLabv32. Auto Annotation tool using Salient Feature detector integration for quicker annotations3. Built Augmentation pipeline to enhance the inhouse product

India

1. Deployed solution to detect forklift and human in a warehouse to alert person walking in forklift lane • Model used :(lightweight) SSD with mobilenet backbone.• Data was collected onsite, labelled dataset, trained model, benchmarked and deployed until desired accuracy levels were met.• Wrote business logic to check IOU overlap and send alerts to PLC.Other models trained during experimentation: • DEEPSORT tracking with Kalman filter, SSD RESNET 50Result: Model was deployed in production which alerts users ensuring employee safety standards2. Document OCR to match BOM with Packing slips • Table extraction using OpenCV• Text extraction and detection using pymupdf, Tesseract.• Trained custom Font on Tesseract. • Experimented with EasyOCROther projects/POCs : 1. Detect Motorcycle parts lhs/rhs for segregation during assembly : model : RESNET2. Classify parts called "Turning Indexable Inserts" which are similar looking parts : model : Densenet

India

During my internship, i learnt 1. System design : how to build optimized end to end solutions catering the needs of a specific customer2. How deep learning models works in production and train them for better accuracy 3. To prune and quantize models during deployment where h/w has insufficient computation powerTechnology stack / programming / libraries/ frameworks : Python, TensorFlow 1.14, Pandas, Keras, numpy, openCV, PIL, Inceptionv4, FRCNN, mobilenet, Resnet.

Master Of Computer Applications - Mca, Computer Science

Bachelor Of Computer Application (Bca), Computer Science