Silt
In developmentA native macOS app that keeps source material current for LLM-maintained project wikis, then gives an agent the exact changes to incorporate through MCP.
GoRustDioxusSQLiteMCP
Patrick Suijk
I build production AI systems and the tools to make them.
I'm a software engineer and technical lead specializing in production AI systems. At RTS Labs, I build AI products for enterprise clients in areas such as logistics, manufacturing, and finance.
I began as a music producer, spending long hours building tracks in Ableton. Music production and software development both involve assembling components and refining a result through iteration.
I care deeply about system behavior below the application layer. When I work in TypeScript, I want to trace a blocking I/O call through libuv and the kernel, across a container namespace, to the disk controller, then back through callback scheduling in the event loop. Detailed systems knowledge leads to better debugging, architecture, and tradeoff decisions.
The focus on system behavior draws me toward infrastructure and systems engineering. Professionally, I write Go for a production backend. Outside work, I study open-source codebases such as Ollama, reproduce scheduler and concurrency failures while preparing an initial open-source contribution, and build systems-oriented personal projects. Some projects solve personal needs, including Magic: The Gathering applications. Other projects provide deeper study of technical concepts and domains.
I'm pursuing infrastructure and systems engineering roles, especially positions where AI is the product. I am open to remote work and relocation for a suitable role.
My long-term goal is to work at the intersection of AI and systems engineering.
A native macOS app that keeps source material current for LLM-maintained project wikis, then gives an agent the exact changes to incorporate through MCP.
A Go agent harness with minimal framework dependencies, supporting tool-use loops, session management, memory, process lifecycle, and file system operations.
A type-safe Python LLM inference library with composable step-function pipelines, Pydantic-based tools, and optional thread persistence.
A Python document-extraction library and full-stack configuration application, moving extraction setup from code to an interface.
Natural-language search for Magic: The Gathering cards. Describe what you're looking for in plain English and receive matching cards.
An MTG companion platform for game tracking, deck and playgroup management, iOS life counting, and personal play-history analytics. I build the backend and web app.
I investigated request-queue behavior in Ollama's model scheduler. The investigation reproduced an issue where requests for already-loaded models bypass the queue-depth limit. I documented the root cause and a two-queue diagnosis for the maintainers.
I am beginning to attend Kubernetes SIG Node meetings to learn the project's architecture and design-review practices. No contribution has landed yet.
SIG Node is a strong fit because the group works on the systems layer immediately below application code. Kubelet and related components manage pod lifecycle, device allocation, topology, cgroups, and runtime isolation. I want deeper experience with the production use of namespaces, cgroups, and seccomp.
The work connects to problems already present in my personal projects. Golem requires decisions around subprocess lifecycle, session state, and permissions. Ollama scheduler investigation involved queues, admission control, and memory pressure. Kubernetes handles production versions of those concerns, especially when GPU allocation must account for hardware topology.
Production systems for enterprise clients in logistics, manufacturing, and private equity. I built each system as the sole developer unless noted. Industries appear by name; client identities remain confidential.
Problem: A large field-service operation needed to coordinate technician schedules and routes using data from several internal systems.
Solution: I contribute to an established production platform with a Go backend and a service architecture integrating external data sources.
Stack: Go, Azure, Snowflake, React.
Problem: Contractors submitted invoices in varied formats, requiring manual entry into the client's internal system.
Solution: I built an AWS Lambda pipeline to extract relevant fields, normalize the data, and send the result to an internal API. The automation reduced manual processing effort by 87%.
Stack: Python, AWS Lambda, S3, Claude.
Problem: A manufacturing client had a large archive of PDFs and legacy Excel files without structured digital query capability.
Solution: I built the backend for an ingestion pipeline and a natural-language-to-SQL agent. The ingestion pipeline processed more than 4,000 PDFs and legacy Excel files, extracted structured records, and populated a Microsoft SQL Server database. The agent lets client staff query historical data in plain English. Each query follows an iterative process: generate SQL, evaluate the results, and revise the SQL when results are insufficient.
Stack: Python, FastAPI, MS SQL Server, Azure, Gemini.
Problem: A private equity firm collected economic metrics from portfolio-company vendors, each with a different file format.
Solution: I built a pipeline to parse source files, extract key metrics, and output a consistent schema. The automation reduced manual processing time by 33%.
Stack: Python, Azure Functions, Claude, Gemini.
The best way to reach me is email. I'm looking for infrastructure and systems engineering roles, especially where AI is the product.