Resume

James Pupalaikis

Machine learning engineer at IBM. B.S. Physics, Carnegie Mellon.

Select a role to expand it

Work experience

Machine Learning Engineer

IBM, New York | August 2022 - Present Increasing responsibility over time in designing and implementing data and AI solutions for clients, tallying $1 million+ in scope and securing over $10 million in continuation and industrialization signings.

Assignment highlights

  • Managed teams of up to 7 developers, both onshore and offshore, to handle multiple data migrations of 50,000+ data assets to target data environments.
  • Spearheaded a team of developers to build 5 small-scale proof-of-concept pilots across different GenAI and LLM oriented use cases, resulting in a large-scale contract renewal for industrialization.
  • Led development of a real-time distributed data analytics pipeline leveraging LLMs to surface customer and order fulfillment insights for a major food distributor.
  • Worked with a team to improve a client helpline response tree using NLP, machine learning, and other data profiling methods.
  • Developed a production-line anomaly detection deep-learning model for a large food and beverage manufacturer, analyzing a multivariate input of vibrational frequency data, temperature data, and technician feedback.
  • Designed and developed an ETL pipeline using AI, NLP, and other data processing methods to build a knowledge graph of clinical semantics relating chemotherapy interactions to various nutrients and therapeutic medicines.

Computer Systems Validation Engineer

Celgene, New Jersey | February 2021 - July 2022 Systems validation engineer in charge of development lifecycle risk planning and requirements validation.
  • Tested and validated FDA regulations for pharmaceutical prescriber platforms.
  • Supported lifecycle planning and management for pharmaceutical portals.
  • Planned and assessed project risks and mitigation strategies for highly regulated healthcare software.

Research Assistant

Carnegie Mellon Physics, Pittsburgh | January 2019 - May 2020 Biophysics research assistant doing image processing and physical data modelling experiments.
  • Collaborated with the lab group on experiment design and procedures involving differential interference imaging of blood cell and bacterial membrane fluctuations.
  • Used image and video processing techniques with deep learning to classify and measure blood cell fluctuations using neural networks.
  • Designed algorithms to determine membrane stiffness from cell fluctuation properties.
  • Designed and executed simulations to determine bilayer physical properties.
  • Presented a research talk as a requirement of an undergraduate research grant program.

Projects

Things built outside of work

Some side projects I am proud of. For a comprehensive list check out my Github.

JamesGAN

Generative adverserial network built from scratch and trained to generate self-portraits PythonTensorflowKerasMatPlotLibPandas

- This was a fun exercise in customn neural network architectures, developed and trained in Python using Keras and Tensorflow
- The outputs are pictures of me, mostly for reasons of practical expedience; Because I built every aspect from the ground up, including curating and processing the training data, this was the easiest thing to use for an output in this case.
- This was made before stable diffusion models became the standard for image generation. I plan to in the future recreate this project using a stable diffusion architecture in PyTorch.

Q-Learning Gin Rummy Player

Gin rummy-playing reinforcement learning model using deep Q learning PythonNumPyMatplotlibTKinterPyTorchSciPy

- I went through a bit of a phase playing too much gin rummy and thought it might be fun to make a reinforcement learning model to learn how to play it
- Implements gin rummy in python using numpy and TKinter for visuals
- Uses deep Q-learning implemented in PyTorch to train a model towards optimal gameplay using a custom reward function
- To do this I had to design a custom parameter set and map these to the possible actions that can be taken each turn.
- Includes a harness to play the game against the bot, and simulate games fto generate training data

HydroAPI

A small local web server on a raspberry pi for monitoring and managing my apartment closet hydroponics system PythonFastAPIasyncioReactJS

- I love to grow plants, especially for culinary purposes but I don't have any outdoor space to keep them. I also don't like to have dirt in my apartment, so the natural solution was to set up a hydroponics garden in my closet.
- This application runs on a raspberry pi 4b in my closet and manages the grow lights, air pumps, and light occlusion status of the garden.
- It also serves to monitor the water level and temperature of the multiple grow buckets, and sends this info plus a live camera feed to a private endpoint on my local wifi network.
- This data can be accessed by me on a custom frontend framework so I can monitor uptime and growth conditions remotely.
- In case you were wondering, I am currently growing 3 kinds of basil, cilantro, thyme, rosemary, and ephedra sinica.
- I have plans to send the feed to a page on this site so users can view the closet but I haven't gotten to it yet.