AI-native builds
Products with an intelligent layer from day one — not a chatbot taped onto the side at the end.
AI-first digital studio / London · everywhere
Nervebox is the brand of Nervebox Ltd, an AI-first web development, automation and digital transformation agency. We build intelligent websites and products, design custom workflows, and automate repetitive tasks — with hybrid, forward-deployed support from requirements gathering to critical deployments and additional support.
Built for the next useful thing
01 / The useful edge
AI is changing the shape of work. We help you change with it — deliberately, commercially, and without turning your business into a science experiment.
Products with an intelligent layer from day one — not a chatbot taped onto the side at the end.
We design custom workflows around how your team actually works, connect the right tools, and automate repetitive tasks and handoffs.
Hybrid, forward-deployed team members work alongside you during requirements gathering, then remain available for critical projects, deployments, and additional support when needed.
02 / The working rhythm
No 80-slide strategy decks. No disappearing into a black box for six months. We work in the open, with a bias toward the next real decision.
A focused working session to separate the real constraint from the loudest symptom.
We design the experience, architecture, and operating rhythm together — so the idea can survive contact with reality.
Small releases, fast feedback, visible momentum. You see the thing becoming useful while we build it.
Your team gets the keys, the context, and a system that can keep changing after we are gone.
03 / Things we have made useful
What we could build for you
A PIM publishing pipeline for a global consumer-electronics brand. An operator uploads SKUs and an Excel copydeck; the tool parses them into one product model, passes every call safely through the corporate network gateway, registers each new model once, then updates features, specifications, SEO and imagery in Adobe Experience Manager on every run — leaving an audit trail and a per-SKU report behind.
Python · Excel (.xls / .xlsx) parsing · Adobe Experience Manager PIM · managed corporate Windows laptop
Stage 0 / Inputs
From a web dashboard with a dry-run mode, the operator feeds in SKUs and content.
SKUs, workbook rows and API JSON all flow into the parser.
Operator console
Sources
0 Inputs
Operator console
Sources
1 Parse & normalise
Unified product model
2 Secure transport
Corporate network gateway
3 Register & update
3a · Direct register — once per new model (if not registered)
3b · Direct update — every run, keeps content in sync
Adobe Experience Manager
4 Outputs
Records
Key: data flow · writes to the CMS · dashed coral = network boundary. No client names, system names or endpoints are shown.
The Tia Tights auto-posting pipeline: a hands-off daily workflow that takes post imagery and videos from Google Drive and publishes them to Facebook and Instagram with fresh, community-focused captions.
Step 01 / Trigger
Fires every morning on a cloud scheduler, so nothing needs to stay switched on.
cron · 0 9 * * * UTCTechnical note: the source schedule is a fixed UTC cron, so local UK time shifts with daylight saving.
Meridian Leads — an intelligent lead management platform for Meridian Partners, transforming how an FMCG consultancy discovers, scores and converts high-value brand partnerships. One place to score brands, sequence outreach and see what is due today.
ATLAS Suite / Full-stack platform / 2024
Designed & built by Nervebox
Lead scoredemo data
A single-operator platform for forex and crypto on QuantConnect, designed for an owner who writes no code. Backtest across parameter sweeps, promote survivors to paper trading, and only then put small real money behind them through OANDA or Kraken.
Status per supplied portfolio: deployed, awaiting first live run
Promotion pathconcept
Five AI tools connected into a single operational brain — automating email, decisions, knowledge, and deployment for a multi-venture business. Built and integrated by Nervebox.
The client was managing eight business ventures from fragmented apps — every decision required context-switching, every email needed manual triage, and nothing remembered anything.
One brain, five capabilities
The problem / scattered tools, disconnected intelligence
48,000+ emails with no automated classification. Every morning began with an hour of inbox archaeology.
Past decisions, client notes, and research scattered across apps with no persistent knowledge layer.
Eight ventures meant eight mental models — switching between them cost hours of cognitive overhead daily.
Individual tools existed, but nothing connected them. No orchestration, no pipeline, no single brain.
Inside the stack / each component configured and integrated
The central AI operating system — deployed on a Hostinger VPS and controlled via Telegram. Hermes autonomously builds apps, manages cron jobs, drafts emails, and delegates to other AI agents.
Example commands: switch to the altitude-ops profile, classify email and save drafts, search the Obsidian vault for a Q4 roadmap, then deploy the portal. This is a replay, not a live terminal or an actual deployment.
The persistent knowledge brain. Every decision, meeting note, research finding, and roadmap is stored in a structured vault that Hermes reads from and writes to in real time.
Illustrative linked-note graph. Notes drift with connections between them; this is not live vault data.
System One AI for structured decisions. Unlike free-text models, Jev returns typed judgments — probabilities, categories, and scores — that feed directly into automated pipelines.
Input — Email Classificationexample
“Hi, following up on the proposal we discussed last week. Could you send the revised pricing by Friday? We'd like to move forward with the Altitude partnership...”
choice → Category: business_followup
A 578-line Python agent running as a daily cron job. It classifies incoming emails via Jev, generates contextual reply drafts with OpenAI, and saves them threaded in Gmail for human review.
Inbox — Automated Classificationexample
Drafts stay in Gmail until a person reviews and sends them.
Workflow panels are illustrative examples from the supplied showcase, not live output or measurements. Paths and URLs are examples, not links.
Tool documentation: Nous Research — Hermes Agent documentation · Obsidian — Create a vault · TypeSafe AI — Decision primitives · Google — Gmail API draft emails. These primary sources explain tool capabilities, not client outcomes.
By the numbers / source-reported
Figures as supplied in the client case study; not live data and not independently verified by this site.
Four weeks to intelligence / iterative weekly sprints
Deployed Hermes Agent on VPS infrastructure. Set up multi-profile system and Telegram integration. Structured the Obsidian vault with venture-specific folders and cross-linking conventions.
Configured Google OAuth with full API scope access. Built the 578-line email response agent with smart filtering, classification pipeline, and draft generation architecture.
Integrated TypeSafe Jev's System One model for structured decisions. Configured noul, choice, and score primitives for email classification, lead scoring, and content evaluation.
Deployed daily cron jobs for autonomous email processing. Finalised multi-profile switching across all eight ventures. End-to-end integration testing and handover.
“The automation stack transformed how we operate. What used to take hours of manual coordination now happens autonomously — and the AI actually understands the context of each venture.”
Ready to automate your operations? We connect AI tools into intelligent systems that work while you sleep. Let's talk about your stack.
Talk to NerveboxBuilt by Nervebox
Codes you can count on
Vocode is an AI-driven consumer service platform built by Nervebox for UK shoppers. Its supplied case study describes an automated voucher verification workflow: ingest codes, test them at checkout without purchasing, classify failures and publish working offers. Merchant submissions, transparent conditions and affiliate disclosure support the consumer experience.
Portfolio demonstration based on the supplied showcase. Animated figures, comparison benchmarks and merchant examples are illustrative or source-reported, not live or independently verified results.
Read the Vocode FAQs and case-study evidence note.
01 / The problem
Most aggregator sites publish codes that don't work. Users waste time trying dead codes at checkout. Vocode fixes this by testing every code before publishing.
The average voucher aggregator publishes codes with a ~26% failure rate. That's 1 in 4 codes that simply don't work — expired, region-locked, or already used.
Vocode takes a fundamentally different approach: every code is tested at a live checkout before it's published. If a code fails, it comes down. If it fails for a specific reason, that reason is published — not hidden.
Comparison figures come from the supplied Vocode showcase; the industry benchmark and claimed success rate are not independently verified here.
02 / The platform
The verified success rate IS the headline. Real data, computed from the database, animated into view. No vanity metrics.
Verified Checkout Success Rate
2,847 codes tested in the last 24 hours
Featured Merchants
Static illustrative preview recreated from the supplied showcase. Counters animate to showcase figures; they are not live Vocode data.
03 / Core interaction
Eight distinct states. Source attribution. Verification status. Failure transparency. Every piece of information a user needs, with none of the deception.
20% Off Everything
Site-wide discount on full-price items. Excludes sale and clearance.
SAVE20NOWIllustrative example code
Every code is labelled: merchant-submitted (green), affiliate network (grey), or community-contributed (amber).
Copy button with real clipboard API integration. Visual confirmation on success, graceful fallback on failure.
If a code fails, the reason is published — expired, minimum-spend, new-customers-only, region-locked. Never hidden.
Pass rate from real checkout tests. Colour-coded: green ≥90%, amber ≥75%, red below. Only codes with ≥1 pass published.
04 / How it works
From ingestion to publication in four steps. Every code is tested at a real checkout — no manual curation, no guesswork.
Ingested from affiliate networks (Awin, Rakuten, Impact, CJ), merchant submissions, and community contributions.
Automated engine attempts code at live checkout. Stops when discount applied or rejected. No purchase made.
Failures are classified: expired, minimum-spend, new-customers-only, region-locked. Reasons published, not hidden.
Working codes go live. Auto-hidden below 70% pass rate. Removed after 7 days without a pass. Re-tested every 24h.
05 / For merchants
Merchants claim their page, submit codes directly, and track performance. Full control over their voucher presence.
Verify ownership and take control of your merchant page. Update branding, descriptions, and featured codes.
Submit codes that go straight to the verification engine. Merchant-submitted codes earn a green 'Merchant Verified' badge.
Track code usage, success rates, and customer engagement. See which codes drive the most conversions.
Set expiry dates and conditions. Codes are automatically removed when they expire — no stale listings.
06 / By the numbers · source-reported
On the live platform, every metric is computed from the database and auditable. The figures below are as supplied in the showcase.
Figures as reported in the supplied Vocode showcase — not live data and not independently verified by this site. What do these portfolio figures mean?
07 / Principles
Vocode's product rules (G6) encode a simple idea: never monetise against the user. Every design decision follows from this.
No affiliate cookie overwriting. No undisclosed code suppression. If a better code exists, it's shown — even if it earns less commission.
When a code fails, users see why: expired, minimum-spend, new-customers-only, region-locked. Failure reasons are data, not secrets.
We may earn a commission when you buy via our links — it never changes your price. This is visible on every page, not buried in terms.
Success rates, code counts, test timestamps — all computed from the database in real time on the live platform. Nothing is hardcoded or aspirational there.
“The best AI work does not feel like AI work. It feels like your business finally got out of its own way.”— The Nervebox operating principle

Founder / Senior Product Manager
The person behind Nervebox
I bring 20+ years of turning complex technical requirements into customer-centred digital products — from multi-market e-commerce transformations to enterprise platform integrations for Samsung, NetJets, Unilever, and the UK public sector.
My work spans checkout and payments, AI/ML implementations, platform migrations, and data-driven UX. The focus is always the same: make the technology useful, and the outcome measurable.
Self-reported career outcomes from the supplied founder bio, across previous roles and projects — not guaranteed Nervebox client results. The LinkedIn profile below provides professional context, not independent verification of these figures.
More about my experience on LinkedInStraight answers
What we do, how the tools fit together, how Vocode’s consumer platform works, and where the evidence ends. Numbered markers point to the sources listed below.
Sources reviewed
Nervebox is the brand of Nervebox Ltd, an AI-first web development, automation and digital transformation agency. Nervebox builds websites and digital products, designs custom workflows, connects systems and automates repetitive tasks. Work starts with understanding the requirements and continues through design, development, deployment and handover.
Nervebox designs automation around the way a client team works. Examples in the Nervebox portfolio include PIM publishing, post imagery and video publishing, lead harvesting and nurturing, outreach coordination, email classification and draft generation. Vocode adds a consumer-platform example: voucher ingestion, checkout verification, failure classification and merchant submissions. The tools, approval steps and integration scope are agreed for each project.
Nervebox provides hybrid, forward-deployed team members during requirements gathering. Team members can also make themselves available for critical projects, deployments and additional support. Delivery combines discovery, experience and architecture design, iterative releases and a handover that gives the client team the context to maintain the system.
In the supplied Altitude Group case study, Hermes Agent orchestrates work, Obsidian Vault provides persistent project knowledge and TypeSafe Jev supplies structured decisions. Hermes supports memory, scheduling and delegation; an Obsidian vault stores notes in a folder; TypeSafe AI provides Noul probabilities, Choice categories and Score ratings. These documented capabilities explain the tools, not the client’s reported results.
The Gmail Agent described in Nervebox’s Altitude Group case study classifies email and saves contextual drafts for human review before sending. Google’s Gmail API documents draft creation and sending as separate operations. Automation can therefore prepare a reply without automatically sending it; the approval policy must be configured for the individual workflow.
Sources[4] gmail
Vocode is the AI Driven Consumer Service Platform presented in Nervebox’s Selected Work: a voucher-code experience for UK shoppers built around automated verification and transparency. The supplied case study describes a consumer homepage, searchable merchant offers, code cards with conditions and verification status, an automated checkout-testing pipeline and a merchant portal. The Nervebox page demonstrates these features rather than running live voucher searches or checkout tests.
Sources[7] vocode-case
The supplied Vocode workflow has four stages: ingest codes from affiliate networks, merchant submissions and community contributions; test them at checkout without making a purchase; classify failures; and publish or reject the result. Failure reasons include expired codes, minimum-spend conditions, new-customer restrictions and region locks. The described rules hide codes below a 70% pass rate, remove codes after seven days without a pass and retest every 24 hours. These are supplied design rules, not independently audited operating results.
Sources[7] vocode-case
Vocode’s supplied merchant-portal design lets merchants claim their page, verify ownership, update branding, submit voucher codes directly for testing, review usage and success-rate analytics, and manage expiry dates and conditions. Merchant-submitted codes receive a source label so shoppers can distinguish them from affiliate-network or community contributions. The portal on Nervebox’s portfolio is an illustrative preview and does not submit, verify or save merchant data.
Sources[7] vocode-case
The supplied Vocode technology stack lists Next.js 14, TypeScript, Prisma ORM, Supabase, Tailwind CSS, Cloudflare, Docker, Traefik, PostgreSQL and a progressive web app (PWA). Next.js documentation describes a React framework for full-stack web applications; MDN explains how PWAs use web technologies to provide app-like experiences. Installability, offline support and background operation depend on what has actually been implemented. These references explain the technologies and do not independently verify Vocode’s deployment or features.
The supplied Vocode principles require commission disclosure, visible failure reasons, no affiliate-cookie overwriting and no suppression of a better code simply because it earns less commission. Its disclosure says Vocode may earn a commission when a shopper buys through an affiliate link without changing the shopper’s price. ASA / CAP guidance says affiliate advertising must be obviously identifiable. That guidance supports clear disclosure; it is not evidence of an ASA endorsement, a compliance audit or Vocode’s business results.
No. On Nervebox’s portfolio, Vocode’s 92% success rate, 2,847 tested codes, 200+ merchants and merchant-specific figures are animated examples taken from the supplied showcase, not a live database feed. The comparison with an approximately 74% industry success rate and the market-first wording are also source-reported, not independently verified here. The case study describes the intended verification and reporting experience; these figures should not be read as guaranteed performance or current operating results.
Sources[7] vocode-case
A website can be made easier to crawl and understand through clear answers, descriptive links, accessible text and structured data that matches its content. Google says its existing SEO fundamentals apply to AI Overviews and AI Mode, with no special AI schema required. OpenAI identifies OAI-SearchBot as its search crawler. These foundations support discovery but do not guarantee indexing, rankings or citations.
Nervebox distinguishes supplied case-study figures, illustrative examples and mocked screens from live systems. The Trading Platform portfolio uses dummy data and is not investment advice. Altitude Group figures and its quote are source-reported, not independently verified by this site. Vocode screens and animated figures are illustrative portfolio examples, not live checkout results; figures and market-first claims in its supplied case study are not independently verified here. The founder’s selected career outcomes refer to previous roles and projects, not guaranteed Nervebox client results.
04 / Your move
Bring us the stuck bit, the half-formed idea, or the thing everyone keeps working around. A useful first conversation is on us.