Computer Scientist

Education

ML Operations Engineer @ Linkt AI (December 2023 - Present)

Developed a video-streaming web application, used Next.js, Roboflow and ONNX to run computer vision models on the client side.
Designed REST and WebSocket API’s on the backend, utilized FastAPI and Tornado to handle real-time requests to image to image matching algorithms on the backend.
Fine-tuned DistilBERT to 98% accuracy on clients dataset, replacing a GPT-3 powered classifier, cutting inference cost by 200%.

ML Engineer Intern @ Dekco AI (June 2023 - September 2023)

Deployed a transformer-based autocomplete engine using Python, Docker and AWS SageMaker, to speed up composition of medical reports for clinicians by 60%.
Fine-tuned 200-million parameter LM’s with CUDA, using K-Bit-Training and qLoRA to minimize compute overhead by 98%.
Designed a text parser to ethically extract relevant data from medical reports, creating the training corpus for an intelligent agent.
Developed an AutoML frontend with StreamLit and FastAPI to receive realtime feedback from domain experts and foster domain-specific model alignment.

Deployed a web app using Python, StreamLit Cloud and OpenAI, enabling readers to create AI-generated illustrations for novels.
Won 1st place at the AI ATL hackathon for Best Use of StreamLit, out of 70 competing teams.

Conducted data analysis of the SoccerTrack dataset, developing player trajectory plots, characteristic area analyses + visualizations
Used the time series data to train an LSTM neural network to predict the future positions of soccer players and the ball.