Expectation
Failed.
Results Delivered.
I take the time to understand how you actually work, then build software that fits. Whether it's full-stack development or adding AI, everything I ship is built around your goals and built to last.
Services
From idea to live system, I handle the whole journey.
Full-Stack Development
React, TypeScript, Node, tRPC. Everything's type-safe end to end, so if the API and UI get out of sync, the compiler yells at me before it becomes a problem. I make early architecture calls that still hold up when you hit 1,000 users.
AI Integration
Most of the value isn't in the model itself. It's in the eval harness that tells you if a prompt change helped, the routing layer that picks the right model for each query, and the cost limits that keep your bill from being a nasty surprise.
SaaS Architecture
Multi-tenant from day one. Auth, billing, admin tools, all built to survive that awkward growth spurt from ten paying users to a thousand.
API & Integration
tRPC where I can, REST where I have to, WebSockets only if someone's watching the screen.
Cloud & DevOps
Containerised deployments. CI that catches regressions before they ship. Alerts that distinguish 'check it tomorrow' from 'wake me up now'.
Security & Performance
Row-level security from hour one. CSP headers, rate limiting, all the unglamorous work so you never have to explain a breach to your users, clients, and investors.
Selective freelance
and contract work.
When you work with Error417, you talk directly with the person building your solution.
I only take one or two new projects a quarter. Most of them start with a 2 to 4 week discovery phase, then move into focused build cycles with weekly demos and milestone-based billing.
Every engagement starts with understanding your workflow and a full analysis of your problem, and the architecture comes out of that.
Full-Stack Architecture
React / Node, tRPC, Supabase, Drizzle ORM. End-to-end type safety means the compiler checks your API contract.
Systems Consulting
Performance tuning, security hardening, DevOps and CI/CD. The hard, unglamorous problems that other people tend to avoid.
AI Integration
LLM pipelines, RAG systems, agent design, prompt optimisation. Ships with an eval suite so you can measure if a change helped, not if the demo looks pretty.
The philosophy behind
the error code.
Error417 fires when reality doesn't match expectations. If you've been burned by an agency that promised four weeks and delivered four months, and your problem had to fit their solution, that's the expectation failure the name is pointing at.
Most agencies want to ship fast. Me, I build to last.
The work I take on tends to be complex enough that doing it badly is immediately obvious, and that naturally filters out anyone looking for a quick, low-effort fix.
Three platforms, one studio.
From a complete prompt engineering workbench, to a personal "Life OS" with cross-hub AI insights, to real-time multilingual communication. Every one is shipping or close to it.
The C.R.E.A.T.E. framework, an auto-optimiser that iterates against CSV test cases, a four-level certification system with anti-cheat session hashing, visual agent building, multi-model debate, a moderated marketplace, and competitive seasons. 35+ features, end-to-end type-safe.
Built because most prompt engineering still happens in a Notion doc and a hope.
C.R.E.A.T.E. Optimiser
Four-phase flow, input, clarification, optimisation, output. Voice input via Whisper. The Auto-Optimiser runs your prompt against a CSV test set so you can tell whether a change helped or hurt.
Certification (L1 to L4)
Server-side exam generation, session hashing to make cheating expensive, public certificate verification, sandbox access gated by the right level.
Agent Foundry
Visual no-code builder for autonomous ReAct-loop agents. Drag-and-drop nodes, typed pipelines, full observability with P50/P95/P99 traces.
AI Council & Benchmarks
Parallel multi-model dispatch for side-by-side response comparison, consensus scoring, and LLM-judge benchmarking across 12 standardised tasks.
Marketplace + Seasons
AI-moderated prompt marketplace, competitive seasons with live leaderboards, weekly challenges, peer-review teams, and creator monetisation tiers.
The connections between hubs are the product. Money tracking that notices your spending spike correlates with stress data from the wellness hub. A learning hub that picks language partners from your IoT history. A dual-database architecture with zero-knowledge envelope encryption protecting the sensitive hubs, even I can't read the data inside them.
Currently in beta. The hard problem isn't any single hub; it's getting users to trust the platform with enough data for the cross-hub insights to actually work.
Voice Assistant
The hard part isn't the streaming, it's keeping the per-message cost down without making the assistant feel stateless. My Skills Manager injects only the context relevant to what you actually said.
Money Hub
Zero-knowledge encryption. The vault unlocks client-side; the server only ever sees ciphertext. Budget tracking and insights without trusting me with your transaction history.
Learning + Translation
Two hubs that talk to each other. The system notices which phrases you stumble on in translation and routes them back into your learning queue.
Wellness Hub
Tracks the metrics, correlates them with the rest of the platform, flags patterns the user can act on. Vault-protected; same envelope encryption as Money.
IoT Hub
One action that's good for your comfort, your sleep, and your bill. The cross-hub logic is where this gets interesting.
You speak in your language. The other person hears it in theirs. The whole pipeline, speech recognition, translation, speech synthesis, runs in under 1.8 seconds end to end. Long enough to be real translation; short enough to be a conversation rather than a walkie-talkie.
The hard work is the latency budget. Every component had to come in under its slice, or the call stops feeling like a call.
Sub-1.8s latency
Optimised speech-to-speech pipeline. Every component runs to a per-stage budget; nothing gets to overrun and ruin the call.
Voice-first
Built around natural conversation, not text boxes. Google Cloud TTS for output that sounds like a person, not a phone tree.
Multi-region
Fly.io edge deployment so latency depends on physics, not geography.
Security audited
Full audit complete with 26 findings addressed before opening to users. Hardened auth and data protection from day one.
What I work on when no one's
asking me to.
Research-grade systems engineering. Not products. Not for sale. The kind of work that informs everything else I build.
ChimaeraOS is a freestanding x86-64 operating system written in C, with a local LLM running in-kernel alongside a predictive-coding spiking neural network as its perception layer. No Linux. No userspace. Three actively-learning systems integrated as one.
Transformer inference with adapter learning, a four-layer SNN with spike-timing-dependent plasticity, and a cognition daemon that routes user intent through tools, all on bare metal. A multi-year project asking whether genuine continuous learning is achievable in a single-process kernel.
Tech Stack
Tools I reach for. No hype picks. Every one earns its place because I've shipped something with it.
Transparent rates.
No hidden fees.
Scope defines the final number. Every engagement is quoted in writing before a single line of code is written. POST /contact → 200 OK, usually within a few hours.
Discovery Phase
£2,200 to £3,500System architecture, tech selection, risk mapping, and a delivery roadmap. Output is a written plan you can take to any developer, me, your in-house team, or someone else if the fit isn't right.
Core Build
From £12,500MVP or platform extension, shipped to production with auth, database, CI/CD, testing, and deployment. Billed in milestones (25/50/25) so we both know where we are.
AI Integration
From £3,000LLM pipelines, RAG, prompt optimisation, latency and cost tuning. Ships with an eval set so you can measure what you're getting, not just how it looked in the demo.
Contract / Embedded
£650 to £1,000 / dayDirect collaboration with your team. Architecture guidance, code delivery, and technical leadership. Three-month minimum, because anything shorter rarely gets useful work done.
Final pricing scoped after a free 30-minute alignment call.
Ready to build
something that works?
Tell me what you're building. I'll tell you how I'd ship it.
hello@error417.com30-min alignment call
Free, no pitch. A real conversation about what you're building and whether I'm the right person to help.
Scoped proposal within 48 hours
Written. Scope, milestones, timeline, price. The number on it is the number you'll see on the invoice.
Paid discovery phase to start
If it fits, we begin with a focused discovery phase. You get a delivery roadmap; I get the context to build it right.