Sounak Saha

I’m Sounak Saha, a Computer Science graduate specializing in Artificial Intelligence and Machine Learning, and I have hands-on experience in developing end-to-end AI systems and agentic applications.

I have extensive experience in working with LLMs, RAGs, and multi-agent systems, including developing a startup simulation platform called Launchpad AI, where I designed and optimized asynchronous agent orchestration and semantic search to minimize latency and cost.

I have also worked on multimodal AI applications, including developing a voice-first AI therapy companion with end-to-end speech pipelines and personality-driven interactions.

I have interned at JNU, ERIC Robotics, and ThinkZone, and I have hands-on experience in developing end-to-end AI applications and systems, including facial emotion recognition, object detection, and AI-driven document processing.

I’m excited about developing end-to-end AI applications and systems that not only have technical merit but also have practical and commercial viability, especially in the context of LLMs and agentic applications.

ML Research Intern at Jawaharlal Nehru University (JNU) (2024-07 – 2024-08)

Built and trained a CNN-based facial emotion recognition model using OpenCV preprocessing, implementing data augmentation and evaluation metrics to achieve 0.97+ accuracy.

• Built and trained a CNN-based facial emotion recognition model using OpenCV preprocessing
• Implemented data augmentation and evaluation metrics to achieve 0.97+ accuracy

ML Intern at ERIC Robotics (2025-01 – 2025-09)

Prepared training datasets and trained a YOLOv11 object detection model, evaluating performance using mAP and optimizing inference speed for robotic environments.

• Prepared training datasets and trained a YOLOv11 object detection model
• Evaluated performance using mAP and optimizing inference speed for robotic environments
• Conducted extensive testing and validation of the model using Flask for deployment on robotic platforms
• Optimized the detection pipeline to enhance computational efficiency for deployment in constrained robotic environments

AI Intern at ThinkZone (2025-09 – 2025-12)

Built and deployed image-processing pipelines using Gemini 2.5 Pro, enabling automated extraction, formatting, and structuring of data from complex document images.

• Built and deployed image-processing pipelines using Gemini 2.5 Pro for automated extraction, formatting, and structuring of data from complex document images
• Developed Flask-based backend APIs to integrate AI image-processing results with the front-end application
• Designed and implemented MongoDB logging workflows to store inference outputs, request metadata, and system diagnostics for debugging and auditability

B.Tech in Computer Science – Sai University (2021-09 – 2025-06)