I lead Business Applications & Intelligence at Calibrium. I co-define the company-wide AI strategy, design agentic workflows that automate real business processes, and build the data platforms and DevOps that keep them running.
- AI strategy and architecture. Adoption of LLMs and SLMs across the company, enterprise AI architecture, and the guardrails that make it safe to ship.
- Agentic workflows. Multi-agent systems that take over complex business processes and decisions end to end, not demos.
- AI-ready data platforms. Pipelines and governance built so that models and agents can rely on the data underneath them.
- Document intelligence. AI-powered document processing, data extraction, and workflow automation.
- Platform and DevOps. GitHub, CI/CD, Docker, and the automation that ties applications, data pipelines, and operations together, with an emphasis on reliability, security, and fast iteration.
AI coding agents waste most of their tokens on discovery: grepping hundreds of files, opening ten to confirm three, re-running the full regression suite because nothing maps a change to a test slice. On a 300-file production codebase I fixed that with four structural changes and cut tokens per task by 70 to 85 percent.
| Pillar | What it does |
|---|---|
| Domain modules | One folder per bounded context, public façade only, boundaries enforced by import-linter in CI |
| Module headers | A 7-line manifest at the top of every file (DOMAIN, PURPOSE, PUBLIC, EMITS, DEPENDS, INVARIANTS, SKILLS) so the agent triages from 7 lines, not 600 |
| Slice-by-default testing | One pytest marker per domain and a policy that runs the smallest plausible test surface |
| Context hygiene | Always-loaded instruction files kept to a navigation index, handoffs archived every session |
Read the whitepaper, or skip straight to the installable skills.
Production-tested Agent Skills that install into Claude Code, OpenAI Codex, Cursor, GitHub Copilot, Gemini CLI and any other tool that reads SKILL.md.
- file-headers: the 7-line module header format, a CI validator, and a prefill script that drafts headers for an existing codebase.
- natural-writing-guardrails: strict rules that strip recurring AI-writing patterns from prose without inventing facts.
/plugin marketplace add K451m/K451msSkills # Claude Code
npx skills add K451m/K451msSkills --all # everything else
- How I cut AI coding token consumption by 80%: the three structural changes, with the whitepaper attached.
- A 7-line manifest beats any prompt-engineering trick: why module headers replace docstrings for AI-assisted review, and how 280 of them were drafted by a 150-line script.
- Field guide and runbook (PDF): the same pattern as a step-by-step playbook.
Open an issue on K451msSkills with questions or war stories from applying the pattern to your own codebase, or comment under the LinkedIn posts above.