01 / whoami

Kieran Charnock

AI NATIVE

LOCATION London, UK

FOCUS Service mgmt × AI deployment

NOW BUILDING Model Price Watch · 2026-08

PROJECTS 18

GITHUB ACTIVITY13,333 contributions in the last year

13,333Contributionslast 12 months

89dCurrent streakactive today

312Busiest day17 Jan 2026

2,794PRs merged2,875 opened, last 12 months

Less More

KBRAINmy second brain, live since 2025-06

kBrain

One MCP server that gives every AI tool I use the same prompts, knowledge and personal corpus, so I never have to re-explain who someone is, why a decision was made or how something is wired.

97/100Sessionsstarted from what was stored

1,600+Articlescurated knowledge

95+Promptshow the work gets done

95K+Emailssearchable, indexed locally

Explore kBrain →

How kBrain works: I start with a question or a task; in 97 of 100 sessions kBrain already had what was needed, and when something is missing I add it once and it stays stored

RECENT BUILDS view all 18 →

18/18 LIVE

Planting Planner

A garden designer's border method turned into 23 numbered rules and a scripted checker; an AI drafts the planting plan, the script counts and checks it, and nothing gets ordered until it passes.

PERSONAL INFRA 2026-09

17/18 LIVE

Local Model Auto-Eval

Watches HuggingFace daily for releases that could replace the local models I run on my own hardware, then stops the incumbent and benchmarks a real candidate against my own production tasks before I hear about it.

PERSONAL INFRA 2026-08

16/18 LIVE

Model Price Watch

Watches what every model I run in production costs, runs tiered evals judged by Claude, and only recommends a swap once a real build-gate bake-off proves it's safe. Never switches anything itself.

PERSONAL INFRA 2026-08

THE STACKwhat everything runs on

Everything my projects and home setup run on, grouped by layer.

Models + agentsClaudeClaude CodeClaude Agent SDKCodexQwen 3.6Gemma 4MLXTypeSafe JevOpenRouterAzure OpenAIGPTOpenAIGeminiWorkers AIOpen WebUIHugging FaceArtificial Analysis
Voice + visionOmniVoicePhonon-2WhisperVoxtralGLM-OCRElevenLabsAzure Document Intelligence
Knowledge + datakBrainMCPChromaQwen3-EmbeddingBM25SQLiteFTS5D1KVAzure SQLInfluxDBPrismaSQLAlchemySanity
AppsPythonTypeScriptGoBunFastAPINext.jsReactHonoTailwindVanilla JSTurborepoGrammyLangGraphA2APWAPicaOAuthnext-auth
APIs + servicesMicrosoft GraphTelegramSpotifyStripeLeafletOpenStreetMapOpen-MeteoOpenRouteServiceBrandfetchBright DataRSS
Hosting + infraCloudflare PagesCloudflare WorkersCloudflare TunnelCloudflare AccessCloudflare R2Cloudflare QueuesAzure FunctionsService BusEntra IDBicepVercelDockerOrbStackApple SiliconlaunchdSSHGitGitHub Actions1Password
Observability + qualityLangSmithLangfuseUptime KumaGrafanaCloudflare LogsSonarQubeSnykOSV-ScannerGrype/SyftClaude Code /code-reviewVitestPlaywright
HomeHome AssistantMQTTZigbee2MQTTPhilips HueBLE presenceFrigateUniFi

HOW I BUILD AI-native

I build with AI in the loop; coding agents in the terminal, with kBrain giving them the same prompts and context every session, and MCP servers I've written so they can work directly with my own systems.

Every morning at 05:30 a scheduled agent picks one of my repos and works through the mechanical findings SonarQube has raised on it, limited to an allowlist of rules that are safe to fix this way; a fix only merges once that repo's own build, tests and CI have passed.

SCANself-hosted SonarQube

FIXmechanical smells, capped & allowlisted

VERIFYagainst each repo's own tests

MERGEonly on a green build and CI

Every fix goes through a pull request; bugs and vulnerabilities are triaged for me to review and are never fixed automatically.

Around that I run several scanners side by side, since each one picks up things the others don't:

Static + SASTSonarQube / SonarCloud: code quality, bug patterns, infrastructure-as-code
Dependencies + secretsSnyk: known CVEs, container & IaC, leaked credentials
Local pre-pushOSV-Scanner + Grype/Syft: offline, before anything leaves the machine
AI reviewClaude Code /code-review ultra: logic bugs the scanners miss