Next-generation WSI workbench

A Whole-Slide Imaging (WSI) platform built for computational pathology and microbiology. Gigapixel slide images become viewable within seconds of upload as an event-driven compiler processes them into read-optimized multiscale pyramids. A viewport-based tile service streams only the tiles relevant to the user over WebSocket, providing a fluid, microscope-like experience even on commodity hardware. Sharded NVMe-backed storage with read-only replicas maximizes throughput and availability, while horizontally scaling services orchestrate ingestion, tiling, caching, and delivery. Eosin's scope is intentionally pure: ultra-high-performance WSI infrastructure.

AI-native LMS. Working title.

An institution-scale Learning Management System (LMS) designed around medical curricula. Synapse combines content-aware AI assistance, event-driven workflows, and telecommunications to support both in-person and remote education. A non-AI tooling layer enforces answer correctness via formal logic (Lean + Prolog) which yields direct citations into the curriculum. The Rust + NATS JetStream + Postgres + Kubernetes backend delivers reliable, low-friction operations while respecting institutional data privacy and large-scale concurrency, whether deployed on-premises or in the cloud. A small pilot with medical students and faculty yielded highly positive qualitative feedback, validating both usability and educational value.

Closed-source project — media, details, and demo available on request.

Hybrid neural/keyword search engine

A multimodal search engine that blends neural and "old school" keyword approaches for robustness, predictability, and controllable retrieval. While general purpose, seer specializes in the full spectrum of medical school curriculum assets—slideshows, PDFs, videos, images, Anki decks, and more—via Kafka-backed pipelines and a Rust + Postgres + Kubernetes core. Each phase of the pipeline is tagged with a cryptographically verifiable audit token, providing full lineage and reproducibility across ingestion, processing, and retrieval. Designed for high-throughput, low-latency use by both humans and AI agents. Serves as the principal search backend for Synapse.

Closed-source project — media, details, and demo available on request.

Biomedical research at scale

An early-stage platform for biomedical analysis, with target applications in bioinformatics, biochemistry, and computational pathology. cyto focuses on collaborative, reproducible workflows—wrapping a suite of analysis services and a Postgres-backed data layer into a cohesive, cloud-native environment. Existing lab-oriented tools include an oligonucleotide designer for PCR and a multisequence alignment analyzer. Current efforts focus on a unified data model spanning NCBI, Ensembl, and related databases to complement a unified search engine.

Multiplayer Doom in the browser

A massively scalable platform for playing custom Doom, solo or with friends, directly in the browser. gib.gg combines a comprehensive WAD Browser for a multi-terabyte corpus with Rust- and C-based game services, WebAssembly/WebRTC clients, and NATS JetStream + Postgres + Kubernetes infrastructure to coordinate matchmaking, session management, and content delivery. Source code is open, the experience is frictionless, and no account is required. Based on a fork of Zandronum to satisfy modern multiplayer expectations.

Open-source software lab

An open-source research and engineering lab I founded to explore medical AI, educational software, and creative computing—as well as providing a collaborative environment for mentoring junior engineers. Now a community of over 100 members, Beebles Laboratorium coordinates numerous projects built on shared, enterprise-grade infrastructure with storage, observability, and real-time telecommunications amenities. Current initiatives span scientific computing, language-learning platforms, and interactive media. We run a proving ground for novel system design.

Fast β-VAE for histopathology image analysis

A high-performance variational autoencoder for whole-slide histopathology images. HistoVAE trains on random WSI tiles via OpenSlide and uses a variety of regularization techniques to converge both reliably and quickly enough to support real-time training and human-in-the-loop analysis. Designed as a foundation for large-scale slide embedding, visual search, and downstream ML pipelines across computational pathology datasets.

Soft real-time vector DB engine

A high-performance vector database engine that treats geometry as a first-class concept. TesseraDB emphasizes explicit topology so that spatial relationships are queryable, interpretable, and stable under incremental updates. Its core algorithms are built around topology-aware neighborhoods rather than naive brute-force comparisons. Multiple optimization layers ensure that query and update performance remain within soft real-time constraints.

Scalable live stream infrastructure

Distributed, fault-tolerant live stream syndication infrastructure designed around real operational requirements: resilient ingest, horizontal scale-out, and clear primitives for orchestration and recovery. strim provides the core building blocks for platforms like Synapse and gib.gg—handling RTMP ingest, H.264 transcoding, HLS segment generation, and coordinated CDN storage across a Rust + Kubernetes control plane.

Various medical AI implementations

A collection of machine learning projects spanning time series analysis, computer vision, NLP, and classical ML—foundational work that shaped my focus on medical AI and large-scale systems.

First commit: 21 November 2020.

Physician shadowing, enhanced.

A proof-of-concept platform integrating real-time transcription, medically aware NLP, and automatic reference retrieval to enhance remote physician shadowing. Built during the COVID-19 pandemic to improve remote shadowing experiences, it demonstrates an end-to-end pipeline from live audio to structured educational scaffolding using Go, Flutter, and Kubernetes-backed services.

Various creative content sites

For an unannounced collaboration with my open source lab. A cluster of Alternate Reality Game / creative concept sites—each its own small experiment—backed by shared Rust + kubernetes infrastructure. The goal is to make you smile while quietly proving out unusual interaction patterns at web scale. For many of these sites, generative content is apparent upon page refresh. Enter username root and password password if prompted for login. Enjoy the easter eggs!