I design and ship LLM infrastructure and platform-grade software end to end. Nineteen MIT-licensed open-source repositories spanning a coding agent runner, a multi-provider gateway, a Rust inference server, storage engines, consensus, and WebAssembly sandboxes. Eighty-seven long-form engineering essays. Ready to bring all of that into a team that is shipping something serious.
One engineer. Nineteen public repositories. Zero hidden trade-offs. Read the code before you read the CV.
Tap a row for evidence
Five-point depth on the things I actually ship. Not maxed across the board, because that would not be true. Every row links to a real repository, a whitepaper, or a long-form essay you can read in ten minutes.
14
engines on the failover ladder behind every model call I ship.
~70%
Disk IO budget reclaimed on Supabase page_views in one perf session, 2026-05-28.
1 day
Turnaround on a genuine enquiry. I read every one myself.
I am a UK-based software engineer with around five years of production experience. The work I am drawn to sits at the intersection of LLM infrastructure and platform engineering: model orchestration, multi-provider gateways, durable workflow engines, low-level inference, storage engines, observability, IaC, and the boring-but-important plumbing that lets a small team ship reliably.
Over the last year I published nineteen MIT-licensed open-source repositories under the same engineering bar I would bring to a team. Typed end to end, tested, documented with whitepapers and architecture diagrams, deployable from a fresh clone. I learn well by writing a thing twice: once to get it working, then again to make it readable. The published repository is the second pass.
I think well in writing. The blog at /blog holds eighty-plus essays on LLM infrastructure, platform engineering, observability, and the modern indie-SaaS stack. Hiring managers tell me reading two or three of those is the fastest way to get a sense of how I think.
I am open to startups raising their first or second round, scale-ups in their series-B-to-D phase, and established product companies with a mature engineering culture. Sectors I find most interesting: LLM infrastructure and tooling, developer platforms, systems and storage, fintech, healthtech, and anything with hard real-world constraints.
Six headlines, twelve more below
Six hand-picked repositories that show variety: an AI gateway, a coding agent, an orchestration engine, a SaaS scaffold, a distributed system, an evals runner. Each is MIT-licensed, runs from a fresh clone, and is the kind of thing I would happily walk you through line by line.
Multi-provider OpenAI compatible gateway with 36-engine failover.
Token efficient coding agent runner with persistent project memory.
Durable multi-agent workflow engine with deterministic replay.
Studio SaaS scaffold with auth, billing, admin, and client portal.
Raft KV store with a fault injection harness.
Evals as code, checked into the same repo as the prompts.
Also in the catalogue
All nineteen open-source repositories at /open-source.
Straight from git log. Not picked, not curated, just what the repository says happened.
Three short comparisons. No fluff.
You get me, the engineer who shipped nineteen production repositories. Not whoever the agency happens to have free that week, behind a project manager who cannot answer the technical question.
Permanent full-time PAYE roles only. No consulting, no contracting, no freelance, no agency subcontract. If you are hiring for a real seat on a real team, this page is for you.
The work you commission is the work that ships. Not something queued behind three other accounts at the agency. You get my full attention on your build, every week.
PAYE, full-time
Permanent employment only. No consulting, contracting, freelance, or agency subcontract until at least February 2030. Companies hiring for a real role on a real team are very welcome to reach out.
Individual contributor role on a real engineering team. AI infrastructure, platform, backend, or full stack. I am the person writing the code, not the manager of someone who is.
When the role is specifically LLM infrastructure, agent runners, multi-provider gateways, evals, or real-time voice systems. I have shipped production versions of all of these and the code is open source.
Early team, real funding, named CTO or technical founder I can talk to directly. Comfortable picking up across the stack and owning systems end to end.
Salary, level, and location are part of the conversation. London, Manchester, Edinburgh, Birmingham, Bristol, Cambridge, Leeds, Glasgow, or remote across the UK.
United Kingdom. I work from a quiet home office on a mechanical keyboard and a pair of large monitors. I run my own production on the same Linux box. Happy on site in London for kick-off, design sessions, and milestones. Open to relocating within the UK for the right role.
UK working hours, async first. Deep work in long blocks before lunch, calls batched in the afternoon. I keep a written log of every working day so I can hand context back to my future self. I do not work weekends unless prod is on fire, and I rarely take a holiday without telling you first.
Strong opinions about code review. Every change deserves a description, a motivation, and the receipts. I read every line that lands in main. I prefer a small, boring, observable stack over a fashionable one, and I would rather ship the smaller thing twice than the bigger thing once.
I am comfortable owning a system end to end, schema, services, API surface, deployment, monitoring, and the runbook a teammate will reach for at three in the morning. I am also comfortable being the new person on a team, and shutting up until I understand the codebase.
LLM infrastructure, platform engineering, storage and consensus, observability, and the indie-SaaS stack. Two or three posts is the fastest way to read how I think before a first call.
The questions recruiters and hiring managers ask first.
No. I am open to permanent full-time PAYE roles only. I am not currently offering commercial consulting, contracting, freelance, or paid project work. Companies hiring for a real role are very welcome to get in touch.
Yes. I sign mutual NDAs as a matter of course before sharing anything sensitive. I have my own template if you do not, and I am happy to sign yours after a quick read.
TypeScript, Python, Go, and Rust. Next.js sixteen App Router on the web, Fastify and FastAPI on services, Postgres on Supabase, Drizzle, Redis and BullMQ for queues. Docker, Kubernetes with Helm, Terraform for IaC. Vercel, Cloudflare, and DigitalOcean for hosting. OpenTelemetry, Prometheus, and Grafana for observability. Multi-provider gateway in front of every model call. Full stack visible at sarmalinux.com/technology.
No. PAYE full-time only. I am not taking agency subcontract work, white-label engagements, or any form of paid project work at present.
Primarily remote from the United Kingdom. I am happy to come on-site in London for kick-off, design sessions, and key milestones, and I am open to hybrid for the right role. UK time zone, written-first communication, video calls scheduled in blocks.
Within one working day. I read every enquiry myself. If the fit is obvious I propose a thirty-minute call, otherwise I write back with the honest reason it is not a fit and a referral if I have one.
Yes for the right permanent full-time role. Notice period and exact start date depend on the conversation. I always read every genuine enquiry within a working day.
You get an engineer who already runs production on the stack he writes about, owns the open-source repositories he ships from, and treats engineering as a serious craft. ADRs, runbooks, observability, the lot. No hand-off, no account manager, no rate-card opacity. The person you interview is the person who writes the code.
I am based in the United Kingdom and open to permanent full-time roles across the UK, London, Manchester, Edinburgh, Birmingham, Bristol, Cambridge, Leeds, Glasgow, remote, hybrid, or on-site. Open to relocation within the UK for the right role.
Permanent, full-time employee roles (PAYE). Senior or mid-level individual-contributor positions across AI infrastructure, AI engineering, platform engineering, backend, and full-stack development. Not taking contract or consulting work.
LLM infrastructure and platform engineering. I have built and shipped slipstream (a token-efficient coding agent runner with persistent memory and a live local dashboard), a multi-provider OpenAI-compatible gateway with 36-engine failover and intent-based plugin auto-routing, a durable multi-agent orchestrator with deterministic replay, a sub-second WebRTC voice agent loop, an evals-as-code runner, a minimal LLM inference server in Rust with paged KV-cache and continuous batching, an LSM-tree storage engine in Go, a Raft KV store with a fault-injection harness, a WebAssembly sandbox in Rust with a deny-by-default host ABI, and full IaC for Vercel plus Supabase plus Cloudflare plus DigitalOcean. All open source. All MIT licensed.
Around five years of production software engineering experience.
TypeScript, Python, Go, Rust, Next.js, Node.js, Fastify, tRPC, Postgres, Supabase, Drizzle, Redis, BullMQ, Docker, Kubernetes, Helm, Terraform, Vercel, Cloudflare, DigitalOcean, OpenTelemetry, Prometheus, Grafana, multi-provider LLM gateways, RAG, vector search, MCP.
Yes, every project is open source at github.com/sarmakska. Nineteen production-shaped repositories spanning LLM infrastructure, agent runners, voice loops, storage engines, consensus, and WebAssembly sandboxes. Whitepapers, architecture diagrams, and quick-start guides on each. Long-form engineering writing at sarmalinux.com/blog. Reading two or three of the posts is the fastest way to see how I think.
Email projects@sarmalinux.com with the company, the role, and one sentence on what you are building. I reply to every genuine enquiry. Recruiters welcome if the role is real and the company is named in the first message.
Send a short note with the company, the role, and a sentence on what you are building. I reply to every genuine enquiry. Recruiters welcome if the role is real and the company is named in the first message.