Labs by Diana — Experiments that ship.

Diana's Digital Twin

A multi-interface conversational agent that answers questions about Diana's work, background, and approach — in her voice, across a standalone web landing page, an embeddable widget, and Telegram. Tiered context injection keeps token costs low without sacrificing depth. Built with Python, FastAPI, OpenAI, and Redis.

CV/JD Matcher & Application Tool

An AI-powered job search platform — upload a CV and job description, get dual ATS and recruiter match scores with career archetype classification, category-aware CV generation with diff view, and interview prep packs. A full job tracker manages the application pipeline with logged interview sessions, AI-generated note summarisation, and an insights dashboard. Built with Next.js 16, PostgreSQL, Drizzle ORM, Better Auth, OpenAI, and Upstash Redis.

AI Chat Scheduler

A conversational AI assistant that helps event attendees discover sessions and build conflict-free personal schedules for multi-day events. Built with the Vercel AI SDK and OpenAI.

GEO Audit

A three-tier AI citability audit — submit any URL and get a scored report measuring how well the page performs across AI search engines. Free basic audit covers three metrics; advanced adds eight with competitor gap analysis; commerce scores against six protocol-specific frameworks including Google UCP, OpenAI ACP, and Amazon Buy for Me. Built with React 19, Vite 7, Express.js, OpenAI, Gemini, Anthropic, and PostgreSQL.

Slack ↔ Claude CLI Bridge

An internal Slack bot that routes direct messages and @mentions to Claude — via the local CLI or the Anthropic API — with GPT-powered Gmail digest and team inbox commands. Built with Node.js, Slack Bolt SDK, and the Anthropic, OpenAI, and Gmail APIs.

Diana Ismail — Portfolio

Diana's portfolio — built from scratch to the same standard she applies to client work. AI-powered project summaries, an embedded digital twin, and a CMS the Owner can update without touching code. Not a template. Not outsourced. Built.

IG Autopilot

An automated Instagram content pipeline — write a line of text, pick a template, and the tool renders a pixel-perfect graphic, uploads it to a CDN, and publishes it to Instagram via the Graph API. Ten designed templates across six content pillars, with single-post and carousel support. Built with Node.js, Puppeteer, Cloudflare R2, and the Instagram Graph API.

What I Built Before Karpathy Named It

In April 2026, Andrej Karpathy published a note on "LLM-Wiki" — a system where AI agents maintain a centralised knowledge base. This article documents the same system, built months earlier across six production repositories and twelve agents, for the simplest possible reason: agents kept losing continuity between sessions, and nobody had finished the tools to solve it yet.

Labs

Diana's project showcase — a hand-coded creative layer where the interface itself is the portfolio piece. Every transition, hover state, and idle animation is built from scratch.

Dark Code: What Ships When Nobody's Looking

What runs in production that nobody fully understands — and the three-layer framework built across six repos to eliminate it. Spec-before-code, behavioral manifests, and a comprehension gate that treats "what fails silently?" as a required deliverable, not a review step.

A Playbook for Multi-Project AI Teams

AGENTIC_WORKFLOW // 4_OF_4

A step-by-step practitioner guide for setting up AI agent governance from scratch -- the three-file foundation, the folder structure, the cross-project coordination layer, the five failure modes to watch for, and the health metrics that tell you whether your system is working or quietly degrading.

Designing Rules for AI Agents

AGENTIC_WORKFLOW // 3_OF_4

Every rule you write for an AI agent has a cost. At 3,000 tokens of instructions, model performance starts to degrade -- not from context limits, but from cognitive load. This article covers the information architecture behind rules that actually work: inheritance patterns, override declarations, what to cut, and why "write clean code" is worse than writing nothing at all.

Scaling an Agentic Workflow

AGENTIC_WORKFLOW // 2_OF_4

How the system from Part 1 breaks at 14 projects -- and the three-layer rule cascade, dual CLAUDE.md architecture, and cross-project coordination that fixed it without changing the agents themselves.

Building an Agentic Workflow

AGENTIC_WORKFLOW // 1_OF_4

How Diana's 12-person AI team actually works — session persistence, context management, cross-project state awareness, and the architecture decisions behind a multi-persona Claude Code workflow that builds and ships real products.