James Cleveland

senior full stack engineer

e-mail: jc@blit.cc
github: @radiosilence
location: London, or remote. Happy to be in an office — I like working with people in person, where the flexibility is real.

Polyglot engineer. Writing code for 25 years, professionally for about 20 of them, across commercial frontend, backend, devops, mobile, and embedded. Elixir and Go services, GraphQL gateways, React and Next.js, native Swift and Kotlin modules, Rust tooling, Terraform, Pulumi and Kubernetes. Frankly I'm not super bothered about what language I am using, as long as it has high quality libs and std and doesn't actively get in the way of achieving the goal.

I enjoy greenfield work, or even taking something to the next level, and hard technical problems I can own end to end. I started out freelancing so had to demonstrably own entire products and make sure they worked properly. So, I want to argue about what a thing should be, not just implement tickets.

Communicating complex technical solutions to stakeholders and fellow engineers in a way they can relate to is key to my approach—if people can't understand what's happening, the most appropriate solution is unlikely to be reached. I've mentored many people in my career and it's important to impress upon them that whilst they may be brilliant, it's about getting everyone else on board with what they're thinking and perceiving.

Big on declarative infrastructure using CI/CD and IaC. It is one thing to be able to build something, it's another to be able to reproduce and scale it in a production environment.

AI and agentic coding are tools like any other: worth learning to use expertly, and no substitute for knowing what you are doing. I have had a lot of fun building MCP servers and gateways in Rust and GraphQL to make them better at their job. Anything like this needs a form of critique and we shouldn't lose our heads — but given the right input, the results can be genuinely impressive.

As a natural creative, what drives me is a job where I wake up every day and build something interesting. I have a huge amount of pride in what I do and can achieve, and am known as the person you go to if you need something done properly.

Selected Work

• Architect and lead engineer of a new Elixir service replacing the reviews domain in Fresha's monolith: ~90M rows migrated, production reads taken from 5% to 100% in two days, ~157M requests a week now.
• A WebSocket server running on the handset, written in Java and Swift as React Native modules, because the TV app it drove was confined to a browser context.
• nano-web—in-memory static file server in Rust, 240+ stars, serving this site.
• Got Microsoft to change Azure Policy—it could not express a compliance check Credit Suisse's CSPM needed, so I made the case at their office and they changed the platform. For my client, and for everyone else using it.
• Found the hole in pip—argued in 2013 that it had to verify SSL certificates, at a point where it fetched packages over plain HTTP and checked nothing. Shipped in pip 1.3 as the fix for CVE-2013-1629, credited by name in the release notes.
• An Android app and the whole AWS backend behind it—React Native, CDK, Lambda, DynamoDB, API Gateway—built from scratch for bike delivery drivers.

Recent Work

Senior Full Stack Engineer, Fresha 2025–Present

Key Skills: Elixir, Phoenix, Ecto, OTP, gRPC, Protobuf, GraphQL, TypeScript, Next.js, React, Zod, PostgreSQL, pgbouncer, Kafka, Snowflake, Datadog, Metabase, LiteLLM, GitHub Actions, Docker, Kubernetes

• Reviews service — architect and lead engineer of a new Elixir service taking the reviews domain out of the legacy monolith, on a marketplace where a venue's rating decides how it ranks. Its own Postgres schema and connection pooler, gRPC and protobuf contracts for other services, a GraphQL surface for clients, and the frontend on top. Delivery owner across web, iOS, Android and backend, then production owner through the rollout: a progressive canary across all four components, reads from 5% to 100% in two days, and I carried the pager for it. Runs at around 157M requests a week.
• Migration — around 90M rows moved with the marketplace live. Reads cut over first, then writes, over a sync that ran continuously so the old system stayed authoritative until it didn't, parity monitored throughout and every stage reversible. The new boundary I agreed with around ten teams whose code still read and wrote the old tables. Moving that much out of a live system needs a tool rather than a script, so I built one: restartable partway through with per-partition ETAs, a throttle that read the database's own health and backed off before it could hurt production traffic, a circuit breaker, per-row error isolation so one malformed record couldn't take down a run, and a diff mode dumping only the rows where old and new disagreed. Fed from S3 history, Snowflake dumps and a live Kafka mirror.
• Sync drift — the two systems kept diverging, and the interesting part was why. I traced it to undocumented callers and background jobs still writing the old tables, and to internal support tasks that had been corrupting review data for years. Built repair jobs scoped to the window where the drift happened, targeted repair for individual staff records, and gave the support team replacement tasks to use instead of the ones causing it. Private replies stayed private throughout.
• Attribution — the monolith credited a review to whoever was on the invoice line rather than the person who actually did the work: about 120,000 misattributed all-time, one in 230. The new service attributes from the calendar booking, keeps that attribution once made, records a row per person who worked on the appointment, and shows that a professional has left rather than dropping them silently.
• Events and warehouse — the service's own changes published to Kafka through an outbox, and the monolith's topics consumed to mirror its writes live while it was still the source of truth, behind a kill switch. Dead-letter topics with depth monitoring, and a reply arriving before its parent review warns and skips rather than crashing the consumer. Rating-change events feed marketplace ranking. The new tables streamed into Snowflake through Postgres logical replication and a change-data-capture connector, with poisoned rows excluded so one bad record couldn't stall the pipeline.
• Postgres and operations — a denormalised line-item table so rating aggregates, search facet counts and sorts read from one place instead of joining across the domain. Composite and covering indexes, GIN full-text search over review bodies, index predicates rather than query filters where the predicate was the win, autovacuum on a fixed insert threshold, seven dead or prefix-redundant indexes dropped. Datadog dashboards with I/O attribution, Metabase parity dashboards, paging on error rate and latency. Debugged the service into production through PgBouncer CA trust, IPv6 bind, and a CPU ceiling that turned out to be the bug rather than tight sizing.
• API surfaces — cut the gRPC surface to seven calls, each scoped to what one caller actually needs, with batch reads made O(1) behind a single windowed query. Held the GraphQL side to the company's pagination standard: query cost ceilings, hardened cursor decoding, field-level redaction, role-based authorisation on reply mutations, and nullable root connections so one failing field degrades the page instead of blanking it. Schema drift and colliding migration versions both caught in CI, each opening its own correcting pull request. On the B2C gateway, generated schemas replaced with ones shaped to the domain, eager resolvers made lazy and batched, and a proper deprecation lifecycle run on legacy fields.
• AI replies — the stack end to end: who is entitled to it, the generator, a reply voice built from the business's own description of itself, a moderation pipeline with the strategy chosen by flag, and the worker that publishes. Drafted first and then published or cancelled, so nothing goes out at a customer unseen. The business's remaining balance read live at every billing decision, and running out cancels scheduled replies rather than failing open. Partner side: a replies tab, an enhance action on drafts, and a countdown before a scheduled reply goes out. 2,343 replies published and 129 businesses on full automation in the first weeks.
• Search — rebuilt consumer search across the SPA, the gateway and the search service, and deleted the previous search outright. Type-specific paginated autocomplete, search history as its own service with Redis failures absorbed rather than pushed at the user, server-side spatial clustering streamed during pan and zoom, and distance measured to a venue's actual boundary rather than a single point, weighted differently for professionals than for venues. The search service runs at around 119M requests a week.
• Loyalty — my first project here and the largest consumer release to date: points-based rewards, the rules for which items a reward applies to and who qualifies, tiers, and the wallet, from schema through gateway resolvers to the UI. Led the parts I had context on, and picked up Elixir on the way.
• Shared libraries — fixed correctness and resource bugs in the Elixir libraries every other service depends on. A broker connection and a Redis process leaked on every failed health probe. Two paths were interning atoms from runtime input, which the BEAM never frees, so given enough traffic the node dies. Also the treatment taxonomy the whole marketplace searches against: CLDR and BCP-47 locale handling, correct pluralisation, gettext catalogues and TypeScript codegen, with CI that regenerates it all and opens its own pull request.
• Supply chain — pushed the organisation towards being genuinely harder to attack through its dependencies, rather than just patched: SHA-pinned actions, toolchains pinned through mise, isolated installs with a build-script allowlist, exact pins and registry-only resolution, and registry-fetch-and-execute patterns killed out of the install path. I wrote the standard that carried it, including the first-party carve-out that stopped the policy being unworkable internally.
• Other engineers — 615 pull requests reviewed for other people. Mentored engineers through complex problems, and worked at product level so technical decisions matched what the business needed. Migrated the gateway from Jest to Vitest, wrote custom ESLint rules for the marketplace codebase, and set up an internal Claude plugin marketplace including a skill that takes a ticket through to an opened pull request.

Senior Full Stack Engineer, Apolitical 2024

Key Skills: Next.js, NestJS, React, TypeScript, Kubernetes, Vite, Express, SCSS, GitHub Actions

• Next.js and TypeScript features for a migration onto a new architecture, with the NestJS APIs behind them.
• Maintained legacy React frontends and Express microservices through the migration.
• Debugged performance problems in services running on Kubernetes, and extended the existing GitHub Actions pipelines.

Senior Cloud Native Engineer, EngineerBetter 2022–2024

Key Skills: AWS, Azure, Kubernetes, Terraform, Concourse, Docker, Go, Python, CSPM, Cloud Foundry, BOSH

• Cloud native consultancy: moving enterprise platforms onto declarative infrastructure and continuous deployment, with reproducibility and resistance to drift prioritised over strict GitOps where the two conflicted.
• Implemented Cloud Security Posture Management policy across cloud platforms. Azure Policy was badly out of step with how the rest of Azure worked, its JSON was poorly documented, and it couldn't express something Credit Suisse needed for their CSPM to work at all—I made the case to Microsoft at their Paddington office and they shipped the change to the platform a few weeks later.
• Wrote Python tooling that audited code and deployments across enterprise estates too large to inspect by hand.
• CI in Concourse, GitHub Actions and GitLab, at a scale where the pipeline is a system in its own right.
• Contributed to Kubernetes External Secrets Operator, mostly pairing with and bringing on less experienced engineers, and to Compliance Framework, a verified CSPM auditing tool.

Consultant Full Stack / Mobile Engineer, Superbike Factory (Freelance) 2021–2024

Concurrent with EngineerBetter and ROXi

Key Skills: React Native, TypeScript, AWS CDK, Lambda, DynamoDB, API Gateway, CloudFront, MobX-State-Tree, BitBucket Pipelines

• Built an internal Android app and all of its infrastructure from scratch for bike delivery drivers: job viewing, notes and photo upload, training with quizzes and video, and taking customer payment.
• Greenfield and serverless throughout—CDK, Lambda, DynamoDB, API Gateway, CloudFront—integrating with the existing systems rather than replacing them.
• React Native client with MobX-State-Tree and a thin layer of AWS Amplify.
• BitBucket pipeline deploying the infrastructure, reading CloudFront outputs back out of it and building the app against them: a new environment needs no manual configuration.
• Audited the existing infrastructure code and shipped the security fixes.

Lead Developer, ROXi 2020–2022

Key Skills: Swift, Java, WebSockets, React Native, TypeScript, Astro, React, Node.js, AWS, MobX-State-Tree, Vite

• Companion App in React Native. The TV app ran inside a browser context, so control went over a WebSocket server running on the phone itself, talking to the television directly across the LAN.
• Wrote the native WebSocket transport for both platforms as React Native modules—Java on Android, Swift on iOS, using Grand Central Dispatch to get the threading right.
• Internal curation tooling on MobX-State-Tree, Tailwind and Vite.
• Statically generated e-commerce site with account servicing in Astro, when Astro was new.

Consultant Frontend Developer, Sapien Interactive (Freelance) 2019–2024

Concurrent with ROXi, EngineerBetter and Superbike Factory

Key Skills: React Native, TypeScript, Firebase, MobX-State-Tree, Node.js, WebSockets

• Brought in by a former business partner to build the app for a new venture and to restart an earlier one, in React Native and Firebase.
• Moved the codebase from class components and Redux to functional components with hooks, wrapped in mobx-react observers.
• Came to MobX-State-Tree sceptical, because I preferred the explicit functional immutability I was used to in Redux, and it won me over: observables, mutable-style updates and flows for side effects, with a fraction of the boilerplate.

Senior Mobile Developer, Zopa Financial Services 2018–2020

Key Skills: Swift, Kotlin, React Native, TypeScript, Redux, Java, Kafka, detox

• Led the credit card section of Zopa's app, React Native and Redux.
• Native modules in Swift and Kotlin against Stripe's card issuing APIs while those APIs were new.
• Kept the codebase current, picking up hooks once they made sense for it.
• detox and @testing-library/react-native for coverage.
• Learned the financial products in enough detail to be useful to the analysts and backend engineers, and fixed backend bugs where that was the fastest route.

Open Source

Whether it be handwritten like back in the old days or architected manually and written with AI, I've always been a proponent of having everything I've produced available to anyone with curiosity.

• nano-web Rust · 240+★—in-memory static file server for SPAs and static content. Serves this site. From my cupboard.
• jaritanet TypeScript—my own infrastructure as a single Pulumi program: it provisions a Hetzner VPS, installs k3s on it, reads the kubeconfig back as an output of the same run that consumes it, and deploys into the cluster it just built, so there is no secret round-trip and nothing for a human to rotate. Cilium as the CNI so NetworkPolicies are actually enforced, Traefik terminating Let's Encrypt TLS over DNS-01, and a censorship-resistant proxy layer—Xray VLESS-REALITY, Hysteria2, unbound, tailscale—running as hostNetwork DaemonSets rather than systemd units, so the host itself runs k3s and sshd and nothing else. Xray owns :443 and passes unmatched traffic to Traefik, so the public site and the proxy share a port. It runs this site, Navidrome, and an MCP gateway with Hydra and Postgres behind it, with VictoriaMetrics and Grafana watching all of it. GitHub Actions previews the stack on a pull request and applies it on merge, and a scheduled job tracks upstream component versions and opens the bump itself.
• fastmail-cli Rust · 65+★—CLI and MCP server for Fastmail over JMAP, CardDAV and GraphQL, with attachment text extraction and masked email. Predates the official Fastmail MCP, largely because I knew what I wanted out of it and couldn't be bothered waiting to find out if they'd want the same.
• MCP servers in Rust—tfl-mcp - this is neat because it also wraps up their REST API into a fully associated graph. Bots love it. codeowners-lsp, mcp-gateway, caldav-cli, mainlynorfolk-mcp. All share a GraphQL transport I designed for them: one typed, introspectable graph instead of a sprawl of flat tools, so a model can discover what exists and ask for exactly the fields it needs. It costs far fewer tokens and it fails in ways a model can read.
• koan Rust · 25★—bit-perfect terminal music player: Ratatui TUI, gapless playback, Subsonic and Navidrome streaming, ReplayGain, spectrum analyser.
• watchwoman Rust—a drop-in watchman replacement that doesn't eat your RAM.
• blit.cc Rust—this site. A static site generator with a content-hashed asset pipeline that fails the build on an unreferenced or hand-written path, and askama_gettext, a gettext implementation for Askama covering 36 locales with CLDR plural rules, checked against CLDR at build time so a catalogue can't disagree with it silently. Nothing reaches the browser but HTML, CSS and a font—the locale picker is command/commandfor and a native <dialog>.
• pip—opened #789 in 2013 arguing that pip had to verify SSL certificates, at a point where it fetched packages over plain HTTP and checked nothing. Shipped in pip 1.3 as the fix for CVE-2013-1629, credited by name in the release notes.
• Contributions elsewhere—TanStack Router (static prerendering fix, and docs), Django REST Framework (timedelta support in the JSON encoder), git-absorb (darwin arm64 build target), react-native-webview, ops (unikernel packaging fixes), go-buildpack (take the Go version from go.mod), icu_ex (compact notation and percent styles for Elixir number formatting), sorl-thumbnail, bowser.
• Earlier—xr 440+★, Ham 380+★, a PHP microframework from when that was a reasonable thing to write, subdown 19★, servers.py 13★, python-nginx 12★, redux-rx-http 12★.

Skills

Daily—TypeScript, Elixir, Rust, GraphQL, Node.js, PostgreSQL, React, Next.js, Docker, Git, GitHub Actions, Tailwind, CSS, bash/zsh, Linux, agentic AI tooling and MCP.

Strong—Go, Python, React Native, Swift, Kotlin, Java, gRPC and Protobuf, Kubernetes, Terraform, AWS (CDK, Lambda, API Gateway, DynamoDB, S3, CloudFront, Cognito, ECS/Fargate, RDS, IAM, Route53, SQS, SES, CloudWatch), Redis, Zod, Vite, esbuild, bun, Zustand, MobX-State-Tree, Redux, RxJS, WebSockets, Kafka, i18n, TDD/BDD.

Worked with—Astro, NestJS, Express, Django, Flask, Celery, Cython, Twisted, MySQL, MSSQL, MongoDB, CouchDB, Couchbase, Memcached, Pulumi, ArgoCD, Ansible, Azure and Azure Policy, Concourse, CircleCI, BitBucket Pipelines, GitLab CI, Traefik, Nginx, Apache, ZeroMQ, Socket.IO, C#, .NET, C++, C, x86 assembly, Qt, PHP, AngularJS, jQuery, SASS/LESS, Cloud Foundry, BOSH, Mesos/Marathon, unikernels, Vagrant, SVN.

Education

Full disclosure I dropped out of a degree in Computer Science & Cybernetics—University of Reading despite winning various competitions for my work. Wasn't for me.

Who is James?

I don't see programming and computers as simply a job, but part of who I am. I shoot photography, mostly street portraits these days, which started out with urban exploration in Berlin. I love cycling and the freedom it gives people. Music and audio matters to me. I go out of my way to find weird little bands nobody has told me about yet, and I'd rather wander into somewhere and chat to someone than have an algorithm hand it to me. I think technology can still be a force by which you can regain control of your own existence, and people are starting to realise this. So I run my own homelab, music collection...I've built my own audio player etc because I don't think that humans should be at the behest of large corporations for something so close to our hearts. Everyone should have a choice.

I'm a keen follower of current affairs, especially from a technical standpoint.

Less Recent Work

Senior Frontend Developer, On The Dot 2017–2018

Key Skills: React, TypeScript, Redux, redux-observable, Go, Node.js, AWS Lambda, API Gateway, Apigee, Auth0, Swagger

• Built the allocation UI that controllers used to assign deliveries and bookings to couriers.
• Modernised the codebase onto React 16, Redux and redux-observable for side effects.
• Owned authentication (Auth0), authorisation (Lambda and JWT), user management, and API aggregation across Swagger, API Gateway and Apigee.

Lead Frontend Developer, SmartFocus 2015–2017

Key Skills: React, AngularJS, Redux, flux, Node.js, Express, WebSockets, ZeroMQ, Redis, C++, C#, .NET, Qt

• Led engineering across the innovation and frontend teams, building and rebuilding frontend systems and the internal services behind them.
• Architected and built three products—shipped and forthcoming—and mentored the engineers working on them.
• Set patterns and practices the wider technical team adopted.
• Worked across database and system architecture, UX and product design wherever that was what the problem needed.

Lead Frontend Developer, Bootbag 2014–2015

Key Skills: React, flux, WebSockets, CSS, HTML

• Prototyped and built a startup's frontend in React, early enough that most of the patterns didn't exist yet.

Technical Director, Links Creative 2013–2015

Key Skills: Django, PHP, AngularJS, jQuery, Node.js, Express, C#, .NET, Linux, nginx

• Technical director of a small Brighton agency, taking client ideas through to shipped products in Django, AngularJS, jQuery and PHP.

Web Developer, Freelance 2010–2013

Key Skills: PHP, Django, Flask, AngularJS, jQuery, Node.js, Linux, nginx, Apache

• Dropped in at the deep end when I moved to Brighton: learned to network, project manage, and lean on rapidly improving technical skills to meet demand. Where the product instinct came from.

PHP Developer / Sysadmin, The Escape Committee 2007–2009

Key Skills: PHP, MySQL, Gentoo, Linux, Apache, Asterisk

• Web development and systems administration, including working out the Asterisk phone systems.