WPI Computer Science, BS/MS ’29, 4.0 GPA
I build infrastructure for AI tooling, and fix the Linux and Python layers it runs on. My event store for MCP servers, mcp-persist, has 19,000+ downloads and runs in production inside mlrouter, the LLM gateway I'm shipping. My fixes are merged upstream in systemd, CPython and across the AI data stack.
Open to Summer 2027 Software Engineering internships
Durable persistence for MCP servers · PyPI · v2.1.1 · 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: 19,000+ 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")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.
/v1/chat/completions across every major provider. The model prefix selects the backend; auto routes on cost.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.
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.
Every pull request I've had merged into projects other people depend on. States are checked against GitHub daily, and Debug replays what each fix changed.
40 merged across 8 repositories, in C, Rust, Zig and Python.
setpriv accepted any prefix of a Landlock filesystem right, so a typo like fs:remove quietly gave a weaker sandbox than the one asked for. Now rejected.
Forced SipHash's round inline where gcc -O2 declined, making short-key hashing up to 6.6x faster across systemd's hashmaps.
Three fixes on journalctl's per-entry read path, including a UTF-8 decoder that ran twice per byte. Output is byte-for-byte identical, 1.7x to 2.5x faster.
The console-width helper used the permissive UTF-8 decoder, so one invalid cell failed a whole table with a bogus out-of-memory error.
column(1) mis-measured cells after any non-SGR escape sequence; the scan now stops at a real ECMA-48 CSI final byte.
Fixed both overflow guards in the load-average parser: one was inverted and one ran after the shift, so values from 2^53 up wrapped to 0.00 instead of being rejected.
One datagram with a bad checksum, sendable by anyone on the link, disabled resolved's mDNS, LLMNR or stub listener until restart.
A leaked reference let resolved free an mDNS service browser while its maintenance timer still used it: a use-after-free, fixed in one line.
Stopped unbounded buffer growth on truncated compressed streams, and added fuzzers that feed hostile input to every decompressor and to importd's qcow2 path.
ProxmoxVE configs with more than one nameserver left DNS unset at boot; now emits one dracut nameserver= per address, IPv6 bracketed.
Truncating a long multi-byte hostname split a UTF-8 character and panicked, failing afterburn-hostname.service at boot.
Re-landed Anthropic-native /v1/models after a staging revert, so Claude Code's model picker populates through a litellm gateway.
The line re-breaker dropped text before an embedded newline and returned entries spanning lines, which could emit uncommented lines in Varlink IDL output.
An empty long option (--=VALUE) prefix-matched every option, and with only one defined was silently accepted as it. Now rejected.
An NGRAM-indexed contains() with a needle under three characters returned zero rows; it now falls back to a full recheck.
An off-by-one skipped the last offset a two-byte escape sequence can start at, so ellipsize() over-truncated strings containing ESC M.
A newline in an ask-password prompt could inject a Socket= line and send the typed password to another path. Unsafe characters are now refused.
Prompts of 255+ bytes wrapped Plymouth's one-byte length field, so the rest of the prompt was parsed as protocol data. Now rejected with -EMSGSIZE.
A truncated multi-byte character at the end of a string made the line breaker step past the NUL terminator and read out of bounds.
\u escapes accepted UTF-16 surrogates and produced invalid UTF-8 downstream; now rejected, matching the \U form.
hostname_cleanup() stripped only one trailing separator, so foobar-- came back as the invalid hostname foobar-.
MRR fusion averaged only over the rankings a document appeared in, letting a single-system hit outrank a consensus pick.
Fixed incorrect reboot/pending logic in sysupdate when --component= is used. First PR merged into systemd.
Implemented Browser.setDownloadBehavior for CDP-driven file download automation.
Fixed incorrect evaluation of list-element docs in the FTS prefilter walk-the-allowlist branch.
Replaced panic paths in the Bedrock embedding provider with typed error returns across serialization, API, and runtime failure modes.
Fixed opaque AttributeError for bare List/Tuple generics in Pydantic schema conversion, raising a clear TypeError instead.
Enabled Anthropic-native model discovery for Claude Code gateway deployments.
Preserved cache_control headers for OpenAI-compatible custom endpoints, fixing silent cache bypass.
Surfaced max_input_tokens and max_output_tokens on /v1/models for accurate client-side capacity planning.
Critical fix: merge_insert silently dropped matched rows when the leading payload column was all-null.
Stopped a cancelled stream from stranding its max_parallel_requests slot, which had been pinning capped keys at their limit.
Added a proxy-wide setting to disable budget reservation.
Moved embedding search onto a dedicated executor to stop event-loop stalls.
Completed file-input support, from FileList through DOM.setFileInputFiles.
The same DataFusion capability carried into LanceDB's Rust core.
Stopped budget enforcement on model-discovery and info routes.
Build-system fix accepted into the reference Python implementation.
DataFusion expressions for merge-insert predicates in a production columnar format.
W3C File API surface implemented in Zig for an AI-agent headless browser.
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.
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.
Back at WPI for the fall, teaching intro systems as a Peer Learning Assistant, serving as a Community Advisor in residence life, and building RunixOS as an independent study.
Upstream, most of my time now goes into systemd and the Linux userspace around it: a use-after-free in resolved's mDNS path, a password-redirect hole in ask-password, and a journalctl read path that runs 1.7x to 2.5x faster. Over the summer I led development of mlrouter, which runs mcp-persist in production.
The lab next to my desk is still 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. 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.
Everything on this page is a command away. Replay any fix from the ledger with debug, or start with help.
Built by hand: HTML, CSS and zero-dependency JavaScript, with no framework and no build step. The backdrop is a 3D graph of merges drawn on a canvas: main runs into the screen, branches fork off and merge back, and scrolling flies you down it. The terminal reads every answer off this page, and PR states refresh daily through a GitHub Action that commits static JSON, so nothing here calls a third-party API while you read. Type is IBM Plex. Hosted on GitHub Pages. Source