Projects & R&D
A portfolio of implementations, engineering work and research — from industrial automation to agentic engineering and LLM infrastructure.
FuelPOS / EvoFuel — fuel station automation
Fuel retail management platform: POS node, dispenser control, tank gauging, multi-vendor equipment integration. 250+ stations, 16 microservices, 152 tanks monitored.
CNG dispenser technical specification
Detailed integration spec for a CNG dispenser with the Topaz-306BU10 controller within the station control system.
STM32G431 recovery firmware on ESP32-S3
Field recovery and reflashing of peripheral MCUs via an external ESP32-S3 platform, cutting on-site service costs.
CMDB AI Agent — BRD (Adversarial Edition)
Business requirements for a CMDB management AI agent, produced under Builder-Validator and delivered in the client’s corporate document template.
a major oil & gas companyAmmonia plant consumption-coefficient analytics
Consumption-coefficient model and technical report for ammonia production, refined through iterative validator passes.
a chemical production plantMining AI: geological modelling & face mapping
Proposal applying ML to geological modelling and underground face mapping, using iPhone LiDAR capture as a low-cost 3D data source.
CNC G-code optimization — discovery
Discovery questionnaire and framework for a CNC toolpath/G-code optimization engagement: cycle time and tool wear reduction.
CCE — Claude Code Enterprise
Enterprise platform for managing AI-assisted development teams: unified configuration, shared pattern library, metrics and security controls.
DevPulse — engineering team metrics
Engineering process analytics product. Differentiating CTO-level metrics: Bus Factor (knowledge concentration) and Wasted Spend (budget lost to rework and idle time).
Claude Code for legacy codebase management
Research into agentic practices for large legacy codebases: mapping, safe refactoring, documentation recovery, regression control.
Agentic development in air-gapped environments
Consulting package and specification for adopting agentic software development inside an air-gapped network: adversarial validation cycles, READY / NEEDS_WORK / NOT_READY verdicts, an 8-level maturity scale.
a state telecom operatorLLM infrastructure evaluation for sovereign deployments
Comparative evaluation of open models and hardware for large-scale on-premise deployment: sizing, quantization, cost per token, workload fit — up to ~200,000 users.
Self-hosted secrets management evaluation
OpenBao vs Infisical CE comparison for internal infrastructure, focused on audit-log completeness and fitness for critical-infrastructure and sovereign environments.
PostgreSQL cluster hardening (28 databases)
Production cluster security programme: group roles, wave-by-wave closure of public access, elimination of a pg_hba.conf trust vulnerability, full ACL sweep audit.
Proxmox cluster upgrade for CVE-2026-53359
Planned remediation of the Januscape vulnerability on the virtualization cluster, closed out with a Builder-Validator validator report.
Live streaming & video commerce architecture
Video commerce platform spec built around a one-on-one consultation storefront model: low latency, scaling, realtime layer.
OpenRA AI Commander
Research project: an AI commander for a real-time strategy game. Fork of OpenRA-RL, gRPC bridge between engine and agent, a hot-reloadable plan.json planning layer, Builder-Validator applied to game strategy.
Adversarial audits of open-source tooling
An ongoing series of validator-pass reviews of agentic and developer open-source projects: claimed capabilities checked against the actual code.
Napkin-Viz — idea visualization (Dan Roam method)
A skill-based tool: classify the problem by 6W, pick one of six visual frameworks, tune for audience via SQVID, then generate the image.