TECHNICAL LAB

Experiment. Measure. Question.

AI model deployment, voice interfaces, GPU architectures, observability, security and recovery drills. Experiments are not presented as production customer deployments.

The Thinker, reimagined with neural networksTHE THINKER · AI & JUDGEMENT

Documented explorations

Containerised LLMs and translation · Documented experiments

Mistral 7B and NLLB in Docker: model packaging, inference boundaries and data-location trade-offs.

MistralNLLBDocker
Explore ↗
Voice assistant architecture · Design and tests

Wake word, Whisper/faster-whisper speech recognition, Gemma, OpenVoice/MeloTTS/Piper. Latency, permissions and local processing are central challenges.

SpeechLLMPrivacy
Explore ↗
Local GPU, dedicated cloud or API? · Architecture comparison

Evaluation of memory, availability, operational cost, data sensitivity and provider exit options.

SovereigntyGPUCloud
Explore ↗
Backup and restore drills · Validated personal infrastructure

Cold backups, integrity checks, automated validation and functional restoration drills on a personal infrastructure. Not represented as a customer assignment.

ReliabilityRecoveryDocker
Explore ↗
Observability · Engineering practice

Prometheus, Grafana, Thanos, logging, service checks and actionable alerts for applications and infrastructure.

MetricsSLOMonitoring
Explore ↗
Secure delivery · Professional practice

SonarQube, Trivy, CI/CD controls, security awareness workshops and release governance.

DevSecOpsCI/CDSecurity
Explore ↗

What is measured

No benchmark numbers are displayed without reproducible measurements.

  • Model quality and the limits of each scenario.
  • Latency and infrastructure requirements when benchmark data is available.
  • Data sensitivity, location and access controls.
  • Recovery, reversibility and maintainability.