Sounak Sarkar

Key Skills

Software

Encord

Labelbox

Roboflow

Scale AI

SuperAnnotate

Top Subject Matter

No subject matter listed

Top Data Types

Computer Code Programming

Document

Geospatial Tiled Imagery

Image

Video

Top Label Types

Bounding Box

Classification

Data Collection

Fine Tuning

Object Detection

Polygon

Segmentation

Freelancer Overview

I am an AI/ML engineer with hands-on experience in computer vision, deep learning, and data-centric AI, specializing in building robust training data pipelines for object detection, image segmentation, OCR, and intelligent agent systems. My work includes developing and fine-tuning models for tasks such as forest fire detection, tree species classification, animal detection in surveillance footage, and document intelligence using OCR and multimodal retrieval-augmented generation (RAG). I am skilled in data annotation, preprocessing, and quality assurance for large-scale datasets, leveraging tools and frameworks like Python, TensorFlow, PyTorch, OpenCV, YOLO, DeepForest, and LangChain. My experience spans domains including environmental monitoring, recruitment automation, and financial analysis, where I have ensured high-quality labeled data and model performance. I am passionate about collaborating with teams to deliver accurate, scalable AI solutions and continuously enhance my expertise in backend systems and cloud platforms.

Labeling Experience

Tree Crown Segmentation

RoboflowGeospatial Tiled ImageryPolygon

So the project idea was to detect the count tree crowns from a drone which would be ran over a forest and separate each tree crown separately using Polygons. After performing the annotation, I used an image segmentation model on the annotated images and got a mAP of 0.92% with precison 0.88% and recall 0.82%.

So the project idea was to detect the count tree crowns from a drone which would be ran over a forest and separate each tree crown separately using Polygons. After performing the annotation, I used an image segmentation model on the annotated images and got a mAP of 0.92% with precison 0.88% and recall 0.82%.

2025 - 2025

Education

Work History