Hi, I'm Sriram Sohan

I am a passionate Machine Learning Engineer and Software Developer dedicated to creating innovative, scalable solutions. With expertise in leveraging state-of-the-art technologies, I specialize in building intelligent systems that drive real-world impact.

Contact Me
Hero image of me

About

Me sitting with a laptop
  • ML icon

    ML Engineer

    Experienced ML Engineer with a history of designing and implementing multiple high-impact machine learning models.

  • AI Systems icon

    AI Systems Developer

    Expertise in building multimodal AI systems, integrating computer vision, NLP, and foundation models to solve complex problems.

  • Data Engineering icon

    Data Engineer

    Skilled in creating and optimizing data pipelines, feature engineering, and deploying models on cloud platforms like AWS and Azure.

  • Research icon

    AI Researcher

    Conducted research in machine learning, focusing on anomaly detection, clustering, and model optimization techniques.

  • Full Stack icon

    Full Stack Developer

    Experience building scalable MERN stack applications with a focus on performance and usability.

Experience

GraphQL

GraphQL

Python

Python

Pytorch

Pytorch

Tensorflow

Tensorflow

HTML

HTML

CSS

CSS

React

React

Node

Node

MongoDB

MongoDB

  • Capgemini Logo

    Senior Software Engineer, Capgemini

    Sept, 2022 - Present

    • Database Design and handling
    • Reduced load times by 30%
  • Verzeo Logo

    Senior Analyst Intern, Verzeo

    Feb, 2022 - Jun, 2022

    • Data Handling
    • Automated ETL pipelines
  • Verzeo Logo

    Machine Learning Intern, Verzeo

    Jan, 2020 - Mar, 2020

    • Developed an AI chatbot
    • Implemented NLP algorithms

RESUME

Discover my journey through a dynamic blend of technical expertise, innovative projects, and professional experience in software development and data analytics. Click below to explore my detailed resume.

Resume Thumbnail
ResumeDownload PDF

Projects

Image of Tennis Analysis

Tennis Analysis

- Developed an advanced Tennis Analysis System leveraging Ultralytics YOLOv8 for real-time detection of players and tennis balls.
- Implemented object tracking algorithms to follow detected objects across frames, enhancing continuous analysis.
- Trained a custom Convolutional Neural Network (CNN) using PyTorch to detect key points on the tennis court.
- Utilized OpenCV for video processing tasks, including reading, manipulating, and saving video files, to support the overall analysis pipeline.

  • Ultralytics YOLOv8
  • PyTorch
  • OpenCV
Source
Image of Parallelized Image Captioning Pipeline

Parallelized Image Captioning Pipeline

- Built an end-to-end encoder–decoder with Bahdanau attention in PyTorch, training on COCO with DistributedDataParallel (DDP) across 4 GPUs, achieving a 3.8× epoch-time speedup over single-GPU.
- Leveraged Dask to parallelize COCO JSON ingestion on 12 cores, realizing a 5.6× speedup and 0.47 efficiency for data preprocessing.
- Implemented GPU-based beam search (k=3) yielding 0.40 images/sec throughput and qualitative attention heatmaps for model interpretability.
- Evaluated caption quality on 500 validation images: BLEU-1 0.63, BLEU-4 0.25, CIDEr 0.77, demonstrating robust generation performance.
- Visualized and compared SP, DP, MP, DDP, and inference metrics in unified Matplotlib dashboards for comprehensive performance analysis.

  • Parallelization
  • PyTorch
  • Dask
  • Multi-GPU
Source
Image of Finance Analyst RAG

Finance Analyst RAG

- Built a Retrieval-Augmented Generation (RAG) system utilizing agents powered by the Groq model.
- Integrated tools like YFinance for stock analysis, analyst recommendations, and financial data retrieval.
- Enabled dynamic responses combining web search and financial insights with real-time data aggregation.

  • Groq API
  • PyTorch
  • OpenCV
Source
Image of Medical RAG

Medical RAG

- Developed an end-to-end medical chatbot leveraging Retrieval-Augmented Generation (RAG) for accurate and context-aware responses.
- Integrated OpenAI GPT and Pinecone to retrieve and generate medical information.
- Designed using LangChain for modular and flexible chatbot workflows.
- Implemented Flask as a lightweight backend for seamless deployment.
- Enabled secure, real-time interactions with sensitive medical data while maintaining compliance.

  • LangChain
  • Flask
  • Pinecone
  • LLM
Source
Image of DCGAN Image Generation

DCGAN Image Generation

- Implemented a Deep Convolutional GAN (DCGAN) to generate realistic images from the CIFAR-10 dataset.
- Designed the generator and discriminator networks with PyTorch for efficient image synthesis.
- Utilized WandB for logging training metrics, including generator and discriminator loss.
- Saved generated images every 2 epochs and visualized quality improvements during training.
- Explored future enhancements, including conditional GANs (cGAN) and Wasserstein Loss for improved stability.

  • Python
  • PyTorch
  • GAN
  • WandB
Source
Image of Medicine-Delivery-System

Medicine-Delivery-System

Developed a secure and efficient database system for managing real-time prescription orders, patient information, and data visualization.

  • Python
  • SQL
  • Flask
Source
Image of LifeCord:Stem Cell Donation

LifeCord:Stem Cell Donation

Developed LifeCord, a Java Swing application connecting blood cancer patients with stem cell donors. The app manages registration, network administration, and treatment coordination across multiple enterprises and organizations.

  • Java Swing
  • SQL
Source
Image of Portfolio

Portfolio

My Portfolio

  • MongoDB
  • Express
  • React
  • Node
Source
Image of Trouvaille- Travel Website

Trouvaille- Travel Website

Created a comprehensive travel website with features for review management, accommodation booking, and user-friendly interfaces.

  • MongoDB
  • Express
  • Node
  • React
Source