Full-stack and AI case studies Nine projects, problem to result.
Each case study covers the problem, architecture, key decisions and results. Five products are live and four are still in development; each study says how its figures were measured. The work spans legal RAG, telehealth commerce, multi-role fintech, e-commerce, per-user AI model training, field-force distribution, a human-in-the-loop scheduling agent and a document research agent.
READ THE CARDER CASE STUDY →[01]Carder — AI Mini-Site Builder
An AI mini-site builder for local businesses: one prompt becomes a complete site in about 30 seconds. I built the site builder end to end, then automated it with RAG over each user's own uploads, and led the backend and data.
150+ users +40% adoption vs first tracked month 3 metered tiersLaravelNext.jsTypeScriptNode.jsExpress+4
READ THE VOLUMIZE CASE STUDY →[02]Volumize — Telehealth E-Commerce Platform
A telehealth commerce platform connecting medical intake, clinician approval, prescriptions, payments, despatch and subscriptions across customer and operations apps.
250+ subscriptions £31K+ revenue 4 back-office rolesNext.js 16React 19TypeScriptMongoDBMongoose+8
READ THE GOLEGAL CASE STUDY →[03]Golegal — AI Legal Workspace
A consumer legal platform for drafting, review, signing, obligation tracking and access to regulated solicitors. I built five modules end to end.
95% accuracy, eval set −30% manual work, user-reported 50+ usersNest.jsNext.jsPostgreSQLOpenAIAgentic RAG+2
READ THE ALFA CASE STUDY →[04]Alfa — Multi-Role Fintech Platform
[IN DEVELOPMENT]A role-aware fintech platform for personal, business and admin journeys over Python microservices. I own sessions, authorisation, onboarding, verification, document handling and the client data layer.
6 isolated capabilities 2 authorisation layers 4 measures instrumentedPythonFastAPISupabaseGraphQLNext.js 16+7
READ THE WISDOMUP CASE STUDY →[05]WisdomUp — Multi-Service E-Commerce Platform
[IN DEVELOPMENT]A three-service commerce platform — storefront, transactional API and CMS admin — built around a checkout designed for its edge cases, not just its happy path.
3 services 8-layer payment guard 14 settlement testsNext.jsTypeScriptExpress 5MongoDBPayload CMS+4
READ THE MUTERPE CASE STUDY →[06]Muterpe — AI Image Generation SaaS
An AI model-training and image-generation SaaS built from the ground up, with per-user training, streamed generation and usage-based billing.
200+ users $5K+ revenue 99% uptimeNext.jsReactPostgreSQLOpenAIFal.ai+1
READ THE SEETRACK CASE STUDY →[07]SeeTrack — Field-Force & Distribution Platform
A field-force and distribution platform for a distributor selling to retail shops in Pakistan: geofenced shop visits, delivery and cash collection, a warehouse stock ledger and double-entry finance. I built ten of its 31 modules, 334 of its 474 API endpoints, and the controls that keep its stock, cash and access correct.
334 of 474 API endpoints built 119 access errors caught pre-release 1,640 test cases, from noneTypeScriptNode.jsExpressMongoDBMongoose+6
READ THE MAILAGENT CASE STUDY →[08]mailagent — Human-in-the-Loop Scheduling Agent
[IN DEVELOPMENT]An AI scheduling agent that watches a Gmail inbox, turns meeting requests into calendar proposals and books nothing until its owner approves them. On its 14-email golden set it scored 92.9% exact match and found all nine meetings with no false alarms, and in its current build all 77 emails of its injection suite passed, even with a stand-in model that obeys every attack.
92.9% exact-match extraction 77 / 77 injection cases passed $0.08 per 100 emails triagedPythonFastAPILangGraphGeminiPostgreSQL+4
READ THE AGENTIC RAG CASE STUDY →[09]Agentic RAG — Document Research Agent
[IN DEVELOPMENT]A multi-tenant research assistant that answers questions over a team's documents and the web with inline citations, and has a second model verify each answer. On a 110-case benchmark against my first version, the rebuild answered more accurately, 95.3% against 91.1% on content checks, reached first token in half the time and cost 39% less per turn.
95.3% answer correctness, 110-case benchmark −52% time to first token −39% cost per turnNext.js 16TypeScriptLangGraphLangChainQdrant+3
jobkit — AI Job-Application Pipeline
A job post in, a send-ready application out. jobkit reads a post from screenshots or pasted text into a structured record, checks eligibility, scores fit against each requirement and drafts the email only from facts the candidate has approved, starting from their own templates. Drafts the model writes pass a claim-by-claim check against the facts they cite before they can be sent, and every send has a 10-second undo.
Private build, not yet deployed and with no live sends. Its evaluation tooling, with Wilson confidence intervals and McNemar tests between models, is built and waiting for its first live run.
[FROM THE BUILD, 18 OCTOBER 2026]
~2 min
Phone upload to ready kit
* first live run, 14 Oct 2026; target 60 s
1,500+
Automated tests
* 1,522 unit and integration, 20 end-to-end
90%
Coverage gate in CI
* lines and branches, core logic
$25
Monthly spend cap
* new jobs stop at $25; every call at $26
[BUGS FOUND AND FIXED IN BUILD]
A page drawn before the send gate closed kept its controls live, so from another tab, the phone or after Back the same email could be sent again, and an email Gmail had already accepted could be rolled back if the log write failed. The controls now open only while the server confirms that exact kit, an accepted send is kept on its own, and only one send per job and role can be in flight.
One-click send counted as a check of every flag, including a line left for the candidate to write and a subject placeholder. A kit holding either is now refused before it reaches Gmail.
A clear restriction such as remote, US only was cancelled whenever another part of the post left the place open, so roles the candidate could not take were drafted anyway. An open or vague place no longer overrides a clear exclusion.
Need something like one of these built?
Start a conversation →