university of michigan cs '26 · open to full-time roles
I build software for messy, real-world operations — real-time voice AI, internal data platforms, and operational dashboards. My best work happens embedded with a team, turning ambiguous workflows into systems people actually use. Go, TypeScript, and Python.
What I'm building, reading, and watching at the moment.
Tether — gentle voice check-ins for aging parents, built around consent, useful summaries, and concrete caregiver follow-up.
The Count of Monte Cristo — Dumas
the most patient revenge plot ever written.
Where I've worked and studied.
Plymouth, MI · Go · Next.js · PostgreSQL
Built and deployed a full-stack internal data platform (Go/Gin, Next.js, PostgreSQL, Docker) that replaced fragmented spreadsheet workflows for engineering and operations teams. Turned ambiguous operational processes into scoped requirements, then shipped ingestion, validation, and dashboards iterated on stakeholder feedback.
Charlevoix, MI
Coordinated logistics and material scheduling for seven distributed crews across a 50-mile radius. Moved from carpentry into pre-construction planning after identifying workflow bottlenecks, redesigning scheduling to improve cross-team coordination and reduce downtime.
Detroit, MI · CAD
Produced 2D/3D CAD drawings and technical documentation for mechanical systems. First exposure to engineering process — how drawings, specs, and real-world parts have to stay in sync.
Ann Arbor, MI
Coursework: Operating Systems, Embedded Systems, Computer Organization, Artificial Intelligence, Software Engineering, Data Structures & Algorithms, Linear Algebra.
Three I'd show first, each with a full case study on the decisions behind it. Smaller projects and experiments below.
Caregiver-controlled AI voice companion for aging parents.
Built around consent, transcript visibility, and reviewable AI summaries that turn into concrete caregiver follow-ups — not black-box monitoring. Browser voice streams through a Go WebSocket API into Bedrock Nova Sonic, landing in a caregiver dashboard.
An internal data platform that replaced fragmented spreadsheets for engineering & ops.
Worked directly with technical stakeholders to turn messy operational processes into scoped requirements, then shipped modular Go APIs, validation and entity-linking pipelines, and dashboards iterated from real feedback.
A live AIS firehose, turned into analyst triage.
Modeled messy, high-volume vessel data into a Foundry Ontology of vessels and anomaly events, validated the stream with PySpark, and shipped a map-based triage view an analyst could actually work from.
Voice, multimodal, low-level systems, and agent tooling.
Phone-based AI patient simulator: outbound Twilio calls role-play a patient via Nova Sonic, with Gemini flagging bugs in the healthcare line under test.
Jobsite photos + voice → a branded PDF report in under 60 seconds. Team hackathon build — I owned the voice-agent and report-generation pipeline.
An x86_64 microkernel from scratch: buddy allocator, 4-level paging, a preemptive scheduler, capability-based IPC, and a tiny userspace shell.
ECS-style C++17/Lua engine with custom allocators, holding 60 FPS at 10k+ entities under a strict 16ms frame budget.
An early experiment in durable, inspectable memory for coding agents — a SQLite event vault with Claude Code / Codex projections.
Long-form notes on the systems that taught me the most — the decisions, the tradeoffs, and what I'd change next time.
Trust in AI voice systems
Why Tether is built to make AI inspectable, not magical: real-time audio over a Go WebSocket API, transcript-backed summaries, and the product decisions that keep distance care from turning into surveillance.
read case study →Forward-deployed · internal tooling
Embedding with engineering and ops at Quantum Opus: mapping how teams actually moved data, then shipping ingestion, validation, entity-linking, and dashboards that fit the organization instead of fighting it.
read case study →Forward-deployed · Palantir Foundry
The engineering reasoning behind the maritime prototype: modeling messy, high-volume vessel data into an Ontology, validating the stream, and shipping a triage view an analyst could trust.
read case study →What I reach for, roughly in order of confidence.