Collin Velarde · Founder & AI Systems Architect
I spent 20 years reading the job from the field. Now I build source-backed AI systems that can read the job with me.
Construction intelligence
drawings, rooms, and code turned into job contextPrivate AI systems
internal tools built on real operating dataSource-backed answers
you can check where every answer came fromAgent operations
coding agents run like a crew, not a demoFeatured work
Three systems built from the same habit: read the job first, keep the sources visible, and ship the thing that survives real use.
construction intelligence
A construction intelligence system that turns drawings, rooms, code, and field knowledge into source-backed operating context.
Field judgment can be encoded into retrieval, review, and structured workflows without pretending the model is the authority.
retrieval and ETL
A pipeline for transforming drawing sets into structured rooms, systems, tags, retrieval records, and reviewable job context.
The hard part is not parsing a PDF; it is preserving enough field context to know what should be held, verified, or released.
private AI systems
Sanitized CRM, evidence, and operator workflow systems for real business processes.
AI systems are useful when they reduce decision drag without leaking client data or bypassing human approval.
Field notes
Notes on running real work with AI systems — written from the field, not the feed. The first essays land on Substack soon and will connect here.
A field note on using AI to run real operating loops instead of treating software as a demo trick.
Notes on why useful AI for operators starts with sources, caveats, and reviewable outputs.
A public teaching lane for turning drawing-family doctrine into apprentice-facing field guidance.
How I work
The pattern is simple: identify the real operating pressure, preserve source truth, encode the review gate, and keep the release decision human-readable.
Start with how the work is actually run: drawings, rooms, releases, holds, and the early catch points that save the job.
Keep sources, caveats, and review state visible so an answer can be checked before it becomes action.
Structure the work so humans and agents can inspect the same proof, metadata, and release gates.
I don't vibe code.
I run work.
Sources stay visible. Releases get a human sign-off. And my name is on the work — the same as it was in the field.
Workflow audits, private AI systems, construction intelligence pilots, and technical collaboration. If it has to survive a real job, it's my kind of problem.