Systems online · CS @ WPI · 4.0 GPA · BS/MS '29
I'm a computer science student who builds infrastructure for AI tooling. My package mcp-persist has 13,000+ downloads on PyPI and runs in production, including inside mlrouter, the multi-model LLM gateway I'm shipping at mlrouter.com. My pull requests are merged upstream in CPython and across the AI data stack.
Open to Summer 2027 Software Engineering internships
Case study
Durable persistence for MCP servers · PyPI · v1.12.1 · MIT
The problem. The MCP Python SDK ships with an in-memory event store. It works until the first deploy. Restart the process and every client's Last-Event-ID points at history that no longer exists, so resumable SSE streams quietly stop being resumable. Add a second worker and it gets worse: each worker holds its own private history.
The design. mcp-persist replaces that single point of loss with swappable, durable EventStore backends. Redis covers production fleets. SQLite covers single-node deployments. PostgreSQL slots into stacks that already run it. The Redis layout is built for replay. An INCR counter issues monotonic event IDs and the events themselves live in HASHes. A per-stream ZSET supports ranged reads so a reconnecting client can resume from any point. TTLs bound storage. A key_prefix namespace makes one Redis safe for many tenants.
Adoption. Shipped at v1.0.0 and versioned steadily forward through real usage: 8,000+ downloads on PyPI and confirmed production deployments. The test suite runs 300+ async tests against fakeredis and real backends, with CI across Python 3.10 through 3.13.
Positioning. mcp-persist plugs into the official SDK's EventStore interface, so it tracks upstream releases and existing servers adopt it unchanged. Where a server can't be modified at all, the PersistenceProxy adds durability from the outside.
from mcp_persist import with_persistence
mcp = FastMCP("my-server")
app = with_persistence(mcp, backend="redis", url="redis://localhost:6379")Flagship · Platform
One OpenAI-compatible API across every model provider · mlrouter.com Live · entering production
mlrouter is a multi-model LLM gateway: a single OpenAI-compatible endpoint that routes requests across OpenAI, Anthropic, Google, Mistral, HuggingFace, XAI, OpenRouter, Deepseek, and many others. It picks a backend by cost or latency, scores each provider's health, and fails over through circuit breakers and fallback chains when one degrades. Virtual keys carry spend caps; every request is logged. It's live at mlrouter.com and going into production with its first customers. It's also the platform where my open-source packages run as first-class services.
One /v1/chat/completions across every major provider. The model prefix selects the backend; auto routes on cost.
Strategies score providers on price and p50/p95 latency, with automatic context-window escalation when a prompt outgrows a model.
Per-provider health scores, circuit breakers, and fallback chains keep the endpoint answering when an upstream goes bad.
Hashed virtual keys resolve to real provider keys server-side, never exposed, with per-key spend caps enforced in Redis.
Streaming sessions survive a restart and resume mid-stream, powered by my own mcp-persist.
Token counts and migration-safety checks run before a model swap reaches production, powered by TokenDrift.
The parts and the whole. mlrouter ships my open source as production services: mcp-persist for durable sessions, TokenDrift for tokenizer safety, a prompt compressor, and a no-code MCP server builder. The packages are the parts; the platform is the whole.
Shipped
A symbol-aware, self-updating vector index and reference graph over a whole repo, served over MCP, so an agent answers "who calls this / what breaks if I change it" in one call instead of a pile of greps. Built on rag-timetravel; keeps itself fresh with a git pre-push hook.
Time-travel debugger for RAG retrieval pipelines. Pin any past query to the exact index version it ran against, replay retrieval, and diff what changed. Now the indexing core that codebase-rag builds on.
Tokenizer diffing for LLM upgrades. Measures token count shifts, cost deltas, vocabulary changes, and remapped IDs before they hit production.
Upstream
Selected pull requests into projects other people depend on. States are fetched from GitHub daily; the receipts link to the source.
| State | Repo | Contribution | Why it mattered | Date | Simulation |
|---|---|---|---|---|---|
| merged | lightpanda-io/browser | feat(webapi): implement W3C File API | W3C File API surface implemented in Zig for an AI-agent headless browser. | 2026-05-25 | |
| merged | python/cpython | gh-150311: Fix minor issues in configure.ac for the CYGWIN port | Build-system fix accepted into the reference Python implementation. | 2026-05-28 | |
| merged | lance-format/lance | feat(rust): support datafusion expressions for merge insert predicates | DataFusion expressions for merge-insert predicates in a production columnar format. | 2026-05-28 | |
| merged | BerriAI/litellm | fix(proxy): don't enforce budgets on model-discovery / info routes | Stopped budget enforcement on model-discovery and info routes. | 2026-06-02 | |
| merged | lancedb/lancedb | feat(rust): support datafusion expressions for merge insert predicates | The same DataFusion capability carried into LanceDB's Rust core. | 2026-06-03 | |
| merged | lancedb/lancedb | fix(python): run AsyncTable.search embeddings on a dedicated executor | Moved embedding search onto a dedicated executor to stop event-loop stalls. | 2026-06-05 | |
| merged | lightpanda-io/browser | Implement input type=file support (FileList, input.files/value, DOM.setFileInputFiles) | Completed file-input support, from FileList through DOM.setFileInputFiles. | 2026-06-05 | |
| merged | BerriAI/litellm | feat(proxy): add disable_budget_reservation general setting | Added a proxy-wide setting to disable budget reservation. | 2026-06-08 | |
| merged | lancedb/lancedb | fix(python): raise clear TypeError for bare List/Tuple in pydantic schema conversion | Fixed opaque AttributeError for bare List/Tuple generics in Pydantic schema conversion, raising a clear TypeError instead. | 2026-06-09 | |
| merged | lancedb/lancedb | fix(rust): return typed errors instead of panicking in Bedrock embedding path | Replaced panic paths in the Bedrock embedding provider with typed error returns across serialization, API, and runtime failure modes. | 2026-06-09 | |
| merged | BerriAI/litellm | fix(proxy): release max_parallel_requests slot when a stream is cancelled mid-flight | Stopped a cancelled stream from stranding its max_parallel_requests slot, which had been pinning capped keys at their limit. | 2026-06-11 | |
| merged | BerriAI/litellm | feat(proxy): surface max_input_tokens/max_output_tokens on /v1/models | Surfaced max_input_tokens and max_output_tokens on /v1/models for accurate client-side capacity planning. | 2026-06-13 | |
| merged | BerriAI/litellm | feat(proxy): serve Anthropic-native /v1/models for Claude Code gateway discovery | Enabled Anthropic-native model discovery for Claude Code gateway deployments. | 2026-06-13 | |
| merged | lightpanda-io/browser | feat(cdp): implement Browser.setDownloadBehavior file downloads | Implemented Browser.setDownloadBehavior for CDP-driven file download automation. | 2026-06-14 | |
| merged | lance-format/lance | fix: evaluate all list-element docs in FTS prefilter walk-the-allowlist branch | Fixed incorrect evaluation of list-element docs in the FTS prefilter walk-the-allowlist branch. | 2026-06-14 | |
| merged | lance-format/lance | fix: merge_insert silently drops matches when a leading payload column is all-null | Critical fix: merge_insert silently dropped matched rows when the leading payload column was all-null. | 2026-06-14 | |
| merged | systemd/systemd | sysupdate: refuse reboot/pending logic when --component= is used | Fixed incorrect reboot/pending logic in sysupdate when --component= is used. First PR merged into systemd. | 2026-06-15 | |
| merged | BerriAI/litellm | fix(openai): preserve cache_control for openai-compatible custom endpoints | Preserved cache_control headers for OpenAI-compatible custom endpoints, fixing silent cache bypass. | 2026-06-16 |
Recent activity
| State | Repo | Contribution | Date |
|---|
Foundations
Completed the CS:APP attack and bomb labs end to end, including the hidden phase and a full ROP chain. Then went past the assignment and reverse engineered the lab harnesses themselves with objdump and GDB.
Wrote an N-way set-associative cache simulator in C with LRU eviction, then used it to drive a cache-conscious matrix transpose. Valgrind traces guided the access patterns.
Independent study · WPI
A capability-based microkernel. CS3013 Operating Systems independent study, advised by Prof. Craig Wills
RunixOS started as a from-scratch microkernel and is now my formal CS3013 Operating Systems independent study at WPI, advised by Prof. Craig Wills. It re-derives classic process, thread, and synchronization coursework on a kernel with no shared memory and no ambient authority, where every right is an explicit capability.
It's a capability-based, IPC-first microkernel written in Rust for x86_64, booting via UEFI on QEMU. The kernel implements capabilities with attenuation and revocation, a preemptible scheduler, synchronous and asynchronous IPC, and a capability-gated filesystem, all backing an interactive console with tracing and profiling tools. A companion research paper examines what preemption does to capability atomicity, tracing a real TOCTOU vulnerability from discovery through a working fix.
Now
Home for the summer, building in the open: maintaining mcp-persist and landing pull requests across the AI data stack. This summer I'm leading development of mlrouter, an OpenRouter-style agentic platform, which runs mcp-persist in production. The lab next to my desk is a Fedora laptop running a 35B mixture-of-experts model through llama.cpp, wired to a filesystem MCP server, because the fastest way to find what breaks in MCP infrastructure is to live on it. In the fall I'm back at WPI teaching intro systems as a Peer Learning Assistant and serving as a Community Advisor in residence life. Before any of this I spent a decade teaching Shotokan karate as a second-degree black belt, which is where I learned to explain hard things patiently. I'm open to Summer 2027 Software Engineering internships.
Terminal
Query systems data, check server stats, or execute tasks interactively.
Email is fastest. I read everything.
Colophon
Built by hand. Vanilla HTML and CSS with lightweight, zero-dependency JavaScript. No framework, no build step for the pages. PR statuses refresh daily through a GitHub Action that commits static JSON, so nothing on this page depends on a third-party API at view time. Type is IBM Plex. Hosted on GitHub Pages. Source