Go Engineering Consultancy

Pigfox LLC is a Go engineering consultancy, operating since 2005. I'm Peter Sjolin, and I build the backend: distributed services, the queues and pipelines behind them, and the parts that have to keep working when a dependency does not. Everything below is either running on this site or shipped for a client.

The record

More than fifteen years of backend engineering, seven of them writing production Go. The work has been distributed systems and the services between them — gRPC and REST APIs, Kubernetes-based deployments, message queues carrying real volume — and, in two regulated settings, payment processing and healthcare claim adjudication pipelines, where being wrong quietly is the expensive failure. More recently that has extended to blockchain and zero-knowledge integration.

Pigfox LLC is incorporated in Nevada and works with clients wherever they are. The company's operating history and my own engineering record are two different spans; the numbers above are mine.

What I build

  • Backend services in Go. APIs, workers and the plumbing between them — typed, tested, and deployed. Not a prototype handed over with a README.
  • Durable job pipelines. Work that must survive a restart belongs in a queue, not a request. I run Postgres-backed queues with retries, error handlers and a dead-letter path, so a timeout recovers instead of failing a user.
  • Concurrency that behaves under load. Bounded queues, backpressure, cancellation that actually cancels. Most production Go problems are not throughput problems; they are what happens at the edges.
  • Systems integration. Payment providers, third-party APIs, chain RPC, whatever the business already depends on — wired in with the failure modes handled rather than discovered later.
  • Diagnosis of an existing system. Profiling, tracing and reading the code you already have, to say plainly where the time and the risk are.

The evidence

The strongest thing I can offer is not a claim, it is a running system you can open. Every demo below is live, and each one exists because it shows a technique that is hard to describe and easy to see. They are the portfolio — the full set is here.

  • Concurrency and cancellation. Goroutine Visualizer runs real goroutines and real channels in this server and shows what a cancelation cannot reach. Backpressure Lab pushes a bounded queue past its capacity under three overflow policies. pprof Lab shows what a block profile sees that a CPU profile cannot.
  • Distributed state and Kubernetes. Reconcile Lab is a controller loop and a competing writer reaching for the same object, built on the real k8s.io object and error types, showing what a reconciler does when a write is refused.
  • Blockchain and zero-knowledge. ZK Escrow releases funds on a Groth16 proof. ZK KYC Pass proves a credential without revealing an identity. RWA Tokenization gates every transfer on an on-chain whitelist. All live on Base Sepolia.
  • Contract engineering. Gas Optimization races two implementations with a compiler-enforced identical interface. Interest Rate Model is an immutable lending curve read straight off the chain. Oracle Health Explorer names what is actually wrong with a price feed.
  • Verification without a chain. Cosmos Lab replays a signed header and two chained ICS-23 proofs offline, and refuses a single flipped bit.
  • WebAssembly. Orbit Lab propagates a thousand orbital objects a frame in the browser. WASM Keygrind grinds EVM vanity addresses with the network disconnected.

The contracts behind those demos go through one pipeline before anything is deployed: a lint stage, a doctrine gate, an address-checksum check, a size gate, the full test suite, complete line, statement, branch and function coverage, Slither at its lowest severity threshold, and two independent invariant fuzzers whose registered property counts must agree with a declared number. It is a public repository, so you can read the gates rather than take my word for them.

Two specialisms, written up separately

Two parts of this work have their own pages, because they are deep enough to deserve one. Production LLM engineering in Go covers the orchestration layer around a model — queues, approval gates, schema-constrained outputs, cost tiering. Automation covers the unattended pipelines, including the one that publishes this site's blog every day without anyone at the keyboard.

AEO audit + fix

Answer engines — the AI assistants and AI search features that answer a question instead of listing links — read a site differently from a person. They lean on structured data, on what a crawler can reach, and on text that still makes sense when it is lifted out of the page. This is a review of a site against those readers, followed by the fixes, shipped.

The case study is this site

Pigfox.com was scanned on check.aeojs.org and scored 84/100. The weakest category was Schema Presence, at 4/20. The cause was measured before anything was changed: 6 of the 11 crawled pages emitted no JSON-LD at all. After the fix the same scanner scored the site 100/100, with Schema Presence at 20/20.

The structured-data emitter written for that fix is now an open-source Go module, github.com/pigfox/jsonld-go, and it renders the shared structured data on every page of this site.

What the review covers

  • Structured data. Which pages emit JSON-LD, which do not, whether the organization, page and content types are present, and whether every reference in the graph points at something the page actually defines.
  • What a crawler can reach. Sitemaps, robots rules, canonical URLs, and which pages fall inside the first set a scanner reads.
  • Text that survives extraction. Whether a page's key statements still read correctly as a standalone passage, since that is the form an answer engine quotes.

What you get

  • A written report: what was measured, on which pages, and why each finding matters.
  • The fixes, made in your codebase or handed over as a patch, with a test that fails if the structured data regresses.
  • A re-scan on the same scanner after the fix, so the before and after are the same measurement.

What this does not do: it does not promise a ranking, a traffic figure, or that any AI assistant will cite you. Those depend on systems nobody outside them controls. What it does is remove the reasons a machine reader would misread or skip your pages.

Price: Contact for a quote. Ask about an AEO review.

Working together

I work solo and end to end, which suits a defined problem better than an open-ended staffing arrangement: a service to build, a pipeline to make reliable, an existing system to diagnose, or an integration that keeps breaking. Small enough to start quickly, and scoped so you can tell whether it worked.

If that fits something you are trying to get built, get in touch.