profile-pic

Shreyas Somashekar

As an AI/ML Engineer, I developed innovative solutions that deliver measurable impact. I created a Question-Answer generation system using LangChain and FastAPI, cutting 40 hours of manual work each month. I also designed and implemented an MLOps pipeline that enhanced computer vision models, achieving a 6x improvement in performance and a 4x increase in detected classes, ensuring the system scales reliably.

I’m passionate about solving real-world challenges through AI. One of my key projects involved designing a proof of concept for accurate exam assessments. By processing audio and video streams together, I helped improve exam integrity, offering a seamless way to verify both streams concurrently.

In a previous role as a Data Scientist, I led the development of an MLOps pipeline that reduced API calls by 90%, making the system more efficient and enabling real-time updates on training progression. I also applied deflectometry to detect surface defects in production, cutting manual labor time by 50%.

As a mentor, I've had the pleasure of working with interns and junior engineers to help them develop their skills and knowledge. I believe that mentoring is an essential part of any engineer's career, and I'm committed to paying it forward by sharing my expertise with others.

I've also had the opportunity to collaborate with frontend and robotics teams on various projects. This experience has taught me the importance of effective communication and collaboration in achieving project goals. I believe that a strong team is essential for success, and I'm always looking for ways to improve our team's dynamics and productivity.

  • Role

    AI / ML & Deep learning engineer

  • Years of Experience

    6 years

  • Professional Portfolio

    View here

Skillsets

  • Backend
  • Machine Learning
  • Deep Learning
  • MLOps
  • Computer Vision
  • Backend
  • TensorFlow
  • PyTorch
  • NLP
  • Python
  • Machine Learning
  • Deep Learning
  • MLOps
  • Computer Vision
  • Python - 7 Years
  • TensorFlow
  • PyTorch
  • NLP
  • Python
  • Deep Learning
  • MLOps
  • Computer Vision
  • Backend
  • TensorFlow
  • PyTorch
  • NLP
  • Python
  • Deep Learning - 5 Years

Professional Summary

6Years
  • Nov, 2023 - May, 2024 6 months

    Machine Learning Engineer

    Examroom.ai
  • Dec, 2021 - Nov, 20231 yr 11 months

    Machine Learning Engineer

    Lincode.ai
  • Jun, 2018 - Jun, 20191 yr

    Data Scientist

    Huawei Technologies
  • Feb, 2017 - Jan, 2018 11 months

    Computer Vision and Deep learning Engineer

    Cloudadic

Applications & Tools Known

  • icon-tool

    PyTorch

  • icon-tool

    MLflow

  • icon-tool

    Sklearn

  • icon-tool

    Mediapipe

  • icon-tool

    HuggingFace

  • icon-tool

    LangChain

  • icon-tool

    OpenAI

  • icon-tool

    Mistral

  • icon-tool

    Llama

  • icon-tool

    AWS

  • icon-tool

    GCP

  • icon-tool

    Kubernetes

  • icon-tool

    Terraform

  • icon-tool

    Docker

  • icon-tool

    AWS Rekognition

  • icon-tool

    AWS Lambda

  • icon-tool

    Python

  • icon-tool

    Java

  • icon-tool

    C/C++

  • icon-tool

    SQL

  • icon-tool

    PostgreSQL

  • icon-tool

    Redis

  • icon-tool

    MongoDB

  • icon-tool

    Pinecone

  • icon-tool

    FastAPI

  • icon-tool

    Django

  • icon-tool

    Celery

Work History

6Years

Machine Learning Engineer

Examroom.ai
Nov, 2023 - May, 2024 6 months
    Engineered a sophisticated multiple-choice question-answer Generative AI application utilizing LangChain and FastAPI, incorporating CI/CD pipelines to automate testing, validation, while cutting manual question generation time by 40 hours per month. Developed MLOps pipeline resulting in an increase of 4x in the number of classes detected and 6x improvement in model performance for computer vision models utilizing ClearML, AWS S3, lambda and AWS Rekognition. Developed a POC to verify audio transcription with on-screen context during exams by concurrently processing audio stream using OpenAI Whisper and video stream with Mediapipe.

Machine Learning Engineer

Lincode.ai
Dec, 2021 - Nov, 20231 yr 11 months
    Conducted rigorous testing and analysis of popular open-source object detection models, delivering actionable insights that drove the strategic deployment of the superior model; improved overall system efficiency by reducing training time by 15%. Engineered MLOps pipelines for training and inference workflows resulting in a 4x increment in the number of training and inference instances with real time reporting of metrics through server-sent events reducing API calls by 90%. Spearheaded a project to fuse Deflectometry principle and AI solutions with collaborative robots and industrial cameras, producing an advanced inspection solution; this innovation reduced manual labor time by 50% in production lines.

Data Scientist

Huawei Technologies
Jun, 2018 - Jun, 20191 yr
    Curated custom datasets from Wider Face, SCUT-BP500, DEX, SSRNet, and FaceNet to develop an advanced facial beauty prediction model, leveraging attributes such as age, gender, and emotion for enhanced predictive accuracy. Improved performance of object detection algorithm by 12% for video summarization and image aesthetics model for EMUI v9.1. Implemented image and video super resolution algorithms, network pruning algorithms to achieve real time inference, yielding a 3dB improvement.

Computer Vision and Deep learning Engineer

Cloudadic
Feb, 2017 - Jan, 2018 11 months
    Leveraged Google Cloud Vision SDK to streamline text detection and extraction processes, resulting in a 50% decrease in document processing time and a 95% accuracy rate for over 10,000 documents processed monthly. Generated 6.2 million images with text to enhance the development of a proprietary OCR pipeline.

Major Projects

4Projects

LLM as a judge

    Incorporated Llama-3.1 8B model as a judge to evaluate the performance of pretrained Llama-3 1B and deepseek-r1 through custom prompts with scoring rubrics. Implemented a streamlit based data dashboard to visualize the results.

RAG based WhatsApp Chatbot using Langchain, Pinecone, and Twilio

    Built a RAG based WhatsApp chatbot to answers queries from Tesla car manuals and deployed the application on AWS ECS.

LangChain application to transcribe, summarize and translate YouTube videos

    Developed LangChain powered Generative AI solution with OpenAIs GPT- 4 to transcribe, summarize, and translate YouTube videos from URLs enabling users to access video content in a more accessible and efficient manner.

Hand pose classification with Mediapipe

    Created a TensorFlow-based model that achieved real-time hand pose classification with an inference time of 4ms.

Education

  • Masters in Electrical Engineering

    University of Central Florida (2016)

Certifications

  • Aws certified cloud solutions architect

  • Deep learning specialization coursera

  • Natural language processing specialization coursera