CS @ Georgia State (2026) — software engineer working across two threads: on-device AI/ML on Apple's platform (Foundation Models, MLX, agentic systems) and production backend/systems engineering (Kubernetes, async services, concurrent data structures). Recent work includes a neural decoder that beats a classical baseline in quantum error correction, and open pull requests against Apple's own MLX inference framework. Open to Software Engineer roles in AI/ML and backend/infra — internships and new-grad.
- maxwell — a neural decoder for quantum error correction, trained and benchmarked entirely on-device in MLX on Apple Silicon. Beats the industry-standard matching decoder (PyMatching's MWPM) on logical error rate at every physical error rate tested, measured on data the model never trained on.
- Open source on Apple's MLX — active pull requests against
ml-explore/mlx-lm: a quantization-loading fix, a KV-cache checkout/return API for memory safety, and a server crash fix — plus root-cause investigations, each with a reproducible test harness, into a quantized-attention memory regression and a speculative-decoding correctness edge case, both corroborated by other contributors on the thread. - keep — a private, fully on-device macOS assistant (Apple Foundation Models, Vision, Speech) that acts on your files, searches your machine, and describes your screen — zero network calls, nothing leaves the device.
- langchain-apple-foundation-models — the first LangChain provider for Apple's on-device models, published on PyPI, verified against LangChain's own compliance test suite, with its integration docs merged into LangChain's official documentation.
- sentinel — a self-healing Kubernetes operator in Go: watches deployment health, triggers rate-limited rolling restarts, live-verified on a real cluster, backed by a three-layer test suite.
Systems, infra & backend:
| Project | What it does | Stack |
|---|---|---|
| sentinel | Self-healing Kubernetes operator — watches deployment health, triggers rate-limited rolling restarts before it just loops forever | Go · controller-runtime · CRDs · Helm |
| ferroledger | Async double-entry ledger service — hand-rolled optimistic concurrency control, append-only WAL-durable log | Rust · Axum · PostgreSQL |
| corecache | Header-only concurrent cache — real O(1) LRU plus a full Adaptive Replacement Cache, lock-free read path | C++20 |
| torchserve-sentinel | Fine-tuned PII/sensitive-data token classifier behind a batched endpoint, with drift monitoring | Python · DistilBERT · FastAPI · Prometheus/Grafana |
| spendtrack | Native Android spend/receipt tracker, integer-cent accounting | Kotlin · Jetpack Compose · Room |
AI & on-device (Apple Foundation Models / MLX / MCP):
| Project | What it does | Stack |
|---|---|---|
| keep | Private on-device macOS assistant — acts, searches your files, describes your screen, full voice in/out (merges private-agent, stacks, lantern into one app) | Apple Foundation Models · MLX · Vision · Speech · AppleScript |
| maxwell | Neural surface-code decoder for quantum error correction — beats minimum-weight perfect matching, trained and benchmarked entirely on-device | MLX · stim · PyMatching · transformer |
| langchain-apple-foundation-models | LangChain provider for Apple's on-device AI (PyPI); integration docs merged into langchain-ai/docs | Python · LangChain |
| orchard | On-device agentic workflow engine — YAML pipelines compiled to LangGraph | LangGraph · Apple Foundation Models · MLX |
| cadence | Local-first health engine — correlates WHOOP/Apple Health with git activity and AI usage, personal baselines, on-device narration, served via MCP. Listed in awesome-mcp-servers. | MCP · SQLite · OAuth 2.0 · Keychain · Apple Foundation Models |
Applied engineering: a connected four-service backend suite over one domain (airport ops) — aerolane (Spring Boot 3 · Keycloak SSO · PostgreSQL · Docker · K8s), checkride (its QA automation framework — REST Assured · Selenium · Testcontainers · Gatling), tarmac (event-driven incident dispatch — Kafka retry/DLQ · MongoDB), and conveyor (batch ETL — Spring Batch 5 · skip/retry + quarantine). Also built two early AI/cloud projects: switchboard-ai (Twilio Media Streams voice agents on GPT-4o) and finance-rag (RAG over bank statements, Supabase pgvector).
Languages: Java, Python, Swift, Go, Rust, C++, Kotlin Backend: Spring Boot 3, Spring Batch, REST, Kafka, Axum AI/On-device: Apple Foundation Models, MLX, LangGraph, LangChain, MCP Data: PostgreSQL, MongoDB, Flyway, Supabase/pgvector QA: JUnit 5, REST Assured, Selenium, Testcontainers, Gatling, Allure DevOps: Docker, Kubernetes, CI/CD, Git, Prometheus/Grafana