me

Prassanna Ravishankar

Machine Learning since 2011

FullStack Machine Learning engineer

The Full Stack ML engineering journey.

This is a 10+ year journey
I started off as a web designer, studing electronics and signal processing. Web designing got me interested in graphic designing, which exposed me to photoshop. Never before, and after, had anyone been so fascinated by the Gaussian Blur feature. Over the next 5 years, I took an interest in Image processing, took up a masters in Computer Vision, got addicted to Semantic Scene Understanding. Over the following 5 years, I gave up on a PhD in Computer Vision, and got fascinated by real world computer vision problems, productionizing research and deploying machine learning applications.
Today, I can say I can build scalable machine learning applications, from ground up, and implement them end to end, with authentication and monitoring.

I occassionally like to write poetry, take some photos and hike aimlessly

Machine Learning

  • Pytorch
  • Numpy
  • Huggingface
  • Scikit-learn
  • OpenCV
  • DeepSpeed
  • Onnx

Engineering

  • FastAPI
  • Streamlit
  • NetworkX
  • asyncio
  • Dask
  • Prefect

Infrastructure

  • Docker
  • Terraform
  • AWS
  • AWS Sagemaker
  • RESTful API

  1. Machine Learning Platforms Team Lead Currently

    Papercup AI

    Skills: Python, Pytorch, Dask, Data analysis, Python packaging, AWS, Docker, Terraform, FastAPI

    • → Created the ML Platforms team and hired two team members, line managing and leading projects for the team.
    • → Deployed collaborative experiment tracking with ClearML, \textbf{saving \$40k and growing}.
    • → Implemented a queue based process using ClearML to train models on the appropriate GPU generation, \textbf{reducing training time by 50\%}.
    • → Implemented a tool using GPT-4 and langchain to automate data generation to reduce the data acquisition time from \textbf{~4 months} to \textbf{~2 weeks}.
    • → Automated and productionised a \textbf{data preprocessing} workflow using Deepgram, GPT-4, Langchain, Temporal, FastAPI and Streamlit, allowing the asynchronous \textbf{system to serve 1000\% more workflows per day}.
    • → Led the team to implement a production pipeline solving the end to end ML lifecycle, reducing \textbf{lead time by 50\%}. Used Dagster, Airflow, ClearML and CI/CD pipelines.

  2. Senior Machine Learning Engineer Currently

    Zenith Ai (Acquired by Opentrons)

    Skills: Python, Pytorch, Dask, Data analysis, Python packaging, AWS, Docker, Terraform, FastAPI

    • → Joined when the company was founded
    • → Part of the team developing a no code AI platform
    • → Implemented a one click solution to export an AI pipeline into a web application
    • → Envisioned and implemented the first version of a self-service infrastructure for app deployment

  3. Applied researcher

    Anyvision (now Oosto)

    Skills: Python, C++, ONNX, CMake, Docker, Pytorch, OpenCV, Depth Estimation, 3D Geometry, Homography

    • → Developed a Focus of Attention system for a retail, vision based analytics platform using 3D geometry, depth estimation models and Head pose detectors.
    • → Developed proof of concept for defect classification in 48 hours, featured in the internal newsletter and piloted in a major electrical company
    • → Introduced a Benchmark-driven-development research philosophy. Created a grafana based analytics service to establish and monitor object tracking metrics.

  4. Image Processing Consultant @ HP

    Altran (acquired by Capgemini)

    Skills: Python, C++, Perl, Embedded systems, Colour science, Docker, Firmware

    • → Supported the creation of a python based middleware for large format printers.
    • → Implemented sensor based calibration routines for large format and 3D printers.
    • → Implemented internal tools to identify and fix calibration bugs, halving debugging time spent by the R&D team.
    • → Conducted courses in Machine Learning and Blockchains (can be found on GitHub.

  5. PhD Student and teaching assistant

    Computer Vision Center

    Skills: Python, C++, OpenCV, Neural networks, Conditional Random Fields, SVM, Random forests

    • → Implemented a computer vision pipeline for Semantic Segmentation of urban scenes, using Random Forests, Bag of words, Conditional Random Fields and Convolutional Neural Networks
    • → Implemented Slither - A random forest framework, built in C++ and Python, with the following paper.

  6. Full stack ML engineer

    Gade Autonomous Private Systems Limited

    Skills: Python, C++, C#, OpenCV, Kinect SDK, Android

    • → Envisioned and Implemented a Virtual Dressing Room (Smart Mirror) based on Microsoft Kinect, which was piloted in a major clothing store.
    • → Implemented SLAM based mapping algorithm for a shopping mall robot assistant using Scitos G5.

  7. Msc. Computer Vision

    University of Sheffield

    Skills: Python, C#, FPGA, OpenCV, VHDL

    • → Implemented Face recognition algorithms, traffic analysis algorithms and FPGA based cyrptography
    • → Final project was to make a ball collection robot for ping pong balls using CMUCAM3

  8. Bachelor of Engineering in Electronics and Communication

    Manipal Institute of Technology

    Skills: Python, C#, FPGA, OpenCV, VHDL

    • → Courses in Signal processing, Speech processing, Microcontrollers and Microprocessors
    • → Joint computing head for IEEE Manipal student chapter
    • → Head of the Internet Development team and member of National Secretariat at IAESTE IndiaMIT
    • → Internships in mdirect, Tunisia (2009), GTL Infrastructure, Mumbai (2010) and FH Westkueste, Germany (2011)