senior @ Johns Hopkins majoring in computer science, focusing on distributed systems. I get excited about building and investing in data/ml infra, devtools, and African founders building technical companies. previously interned at Microsoft where I helped with building an inference framework for model serving for first-party applications such as Paint, Nvidia where I built dev tools and deep learning infra for robotics, and worked in engineering at multiple startups such as building crypto off/on ramps, ERP solutions for non-profit organizations and as a pm intern for an informal retail growth stage startup. Also, did short stint at twosigma vc mostly focused on crypto(infra) vc, and b2b software infra respectively. would totally love to chat - shoot me a note @ either bnjorog2@jhu.edu or bill@bnjoroge.com or bill@alevelcapital.com(for Hopkins-affiliated founders)
Work
Microsoft — Software Engineering Intern— 05/23 - 08/23
- Wrote a production-grade API that serves a deployed stable diffusion model for use cases such as background and object removal.
- Reduced the API latency and cost by 4x, and increased the throughput by 4x using techniques such as quantization, request batching, converting the Pytorch model to ONNX, caching, and using tools such as Deepspeed.
- Contributed to the content safety toolkit for deploying LLMs.
Tech stack: Python, PyTorch, Azure, ARM templates, FastAPI.
Nvidia — Software Engineering Intern, ML Infra — 05/22 - 08/22
- Trained and deployed(across multiple GPU nodes) an end-to-end rcnn neural network using Pytorch and opencv that did object recognition and instance segmentation on gigabytes of synthetic vision data, demonstrating the project's validity.
- Built middleware that enabled the team to programmatically launch workflows with much better composability, less dependency and better integration(homegrown orchestration tooling)
- Used Dask and PySpark to build pipelines that parallelized multiple workflows(such as synthetic data generation and simulation), reducing total compile time by 20% and supporting 8x more data sizes.
- Reproduced multiple research studies, mostly on embodied AI and computer vision, and presented to the team.
Tech stack: Python, Dask, PySpark, Docker, Jenkins etc.