Original toolbox
The grid from the first version of this site, before most of the above.
60 tools, grouped by area, with what I used each one for and where. Studio and client work is kept generic; the projects and posts go deeper.
Where most of my time goes: services, pipelines and the databases behind them.
Services, pipelines, provisioning engines and scrapers. My default for anything server-side.
Marketplace, commerce, healthcare and booking-agent backends; OpenAPI-documented endpoints; enterprise middleware.
Middleware connecting contact-centre platforms, with retries and circuit breakers.
Worker APIs, React apps and a storefront SDK and CLI.
APIs on Cloudflare Workers: a free-tier storefront and a work-order board.
REST and MongoDB-backed services.
Primary database for the marketplace, commerce platform, booking agent and work-order board.
Interview practice tool and my earliest projects.
Document store for internship services and early projects.
OTP challenges with atomic rate limits and lockouts, plus caching.
ORM for the booking agent and the interview tool.
A streaming transcription pipeline, and real-time voice in the interview tool.
Building things that call models, and things that models call.
Intent classification, structured extraction from websites, and multi-agent flows.
Conversational handlers with tool loops, and day-to-day agentic coding.
A multi-agent booking and discovery assistant with 15+ database-backed tools.
Exposing commerce operations and clinic information to agents, with OAuth and plan-aware tool gating.
Tenant-scoped tool loops, and an eval runner as part of a rebuild plan.
Speech-to-text and text-to-speech for real-time voice interviews.
The glue between platforms, which is most of what makes software useful.
A sub-account per business, WhatsApp senders and an automated account lifecycle.
Embedded signup, business account connection and sender setup.
Inboxes, agent bots, human handoff, SSO login links and automated provisioning.
Checkout sessions, product and price sync, and signed webhooks.
Order creation, payment verification and webhooks for a stays marketplace.
Salesforce, Genesys, Talkdesk and NICE CXone integrations.
HMAC-signed commands with replay guards, and OAuth for agent access.
Availability feeds for calendar sync between a marketplace and its hosts.
A lead pipeline through a service account, keeping attribution columns.
Meta and GA4 server-side events, sent when a lead becomes revenue.
From Lambda fleets to one Worker per tenant.
Lambda, SQS, DynamoDB, Aurora, Cognito, CloudFormation, WAF, CloudWatch, EC2 and S3. A batch job went from 170 s to 40 s on distributed Lambda workers.
Redesigned a BigQuery process from 10 GB to 800 MB of memory.
A Worker per store, signed config in KV, tenant databases in D1, uploads in R2.
Application stacks, migration containers and workers.
Reverse proxying and certificates inside a zero-touch provisioning engine.
Queues, asynchronous routing and dead-letter retries.
CI that validates docs structure, architecture layers and tests.
Servers, shells and everything in between.
Kept separate from the application on purpose.
The single UI, with datasources and dashboards provisioned from files.
Centralised container logs, labelled from Compose metadata.
Metrics from host and container exporters.
Container and host metrics: CPU, memory, disk, network and temperature.
Enough to ship the whole product, not to win design awards.
Admin dashboards, merchant portals and the interview tool.
A customer-facing marketplace and storefronts.
A patient mobile app for ordering medicines and booking appointments.
This site, storefront themes and an agency marketing site.
Build tooling for most React apps.
Styling a storefront on Workers.
The work-order board UI.
UI and state for the interview tool.
Data fetching and caching in the work-order board.
An installable merchant app for hosts.
The layout of this site, unchanged since the first version.
Making pages findable by people and by answer engines.
Measuring what actually earns clicks, and where pages compete with each other.
Auditing pages for citation and answer extraction, and writing the tools to do it.
Structured data that matches the visible content.
A markdown index of a site for AI crawlers.
The habits around the tools.
A short decision record for each architectural choice, so scope arguments become edits.
Shared fixtures asserted from both sides of a service boundary.
Short agent-readable maps, checked structure and graded quality per domain.
Agentic coding, audits and repo-level workflows.
The grid from the first version of this site, before most of the above.
