Home Pricing Case study FAQ Contact us
Case study · The proof of concept

Built inside Calichi Design Group.
Ready to deploy inside your firm.

Six months. 600+ hours. 15+ specialist agents handling fire flow reports, proposals, contract review, plan QC, utility research, and more — in production today, on real projects with PE-stamped deliverables. This is the build that proved Firma AI’s architecture works end-to-end inside a live engineering firm.

A live engineering firm. A licensed engineer. A real production system.

Reco Prianto is a licensed civil engineer with 26 years of industry experience and the founder of Calichi Design Group and Firma AI. He spent six months building the system inside Calichi — 600+ hours, 15+ specialist agents, a production deployment handling the firm’s full business. Not a prototype. Real projects. Real deadlines. A PE stamp on every output. Firma AI exists to deploy that same architecture inside yours.

Founder
Reco Prianto, P.E.
26 years in civil engineering · AI builder
Founder, Calichi Design Group (4 offices, CA / OR / HI)
P.E. licensed across 8 states · worked in 42
Built Calichi’s production AI over 6 months
15+ specialist agents in production today
Founder, Firma AI
Reco leads discovery and architecture directly — not through sales reps or account managers.

Licensed engineers spending the majority of their time on production work — reports, proposals, compliance documents, contract review — the same deliverables with the same structure on every project. High-value staff on low-variance tasks.

What we discovered

The bottlenecks

The discovery process surfaced the workflows where licensed engineers were spending time on work that didn’t require a licensed engineer.

  • Fire flow reports: 16–24 hrs each, same calculation pattern every time
  • SWPPP packages: repetitive compliance checklists and permit language
  • Proposals: 80% identical structure across jobs, written from scratch each time
  • Contract review: same risk flags flagged on every EJCDC/AIA contract
  • Utility research: hours of public records lookups producing 3-page summaries
  • Plan QC: manual cross-discipline checklists run by the same engineer every time
What we built

The system

Each bottleneck got its own agent. The orchestrator (the firm’s AI) routes incoming work, manages the fleet, and returns the deliverable to the project folder.

  • Named AI with Calichi’s identity, voice, and standards
  • 15 specialist agents covering technical work and business ops
  • Email bridge: any authorized employee emails the AI — gets back a finished deliverable, no direct system access needed
  • Council for judgment calls on scope, risk, and contract positions
  • QC gate that checks every output before it leaves the system
  • Full firm config: 4 offices, 8-state PE coverage, templates, routing
Production-measured time savings

These numbers are measured in production at Calichi. Not extrapolations, not best cases — the actual time the system takes vs. the manual baseline.

Workflow Manual With AI Saved
Fire flow report 16–24 hrs 2 hrs 14–22 hrs
Engineering proposal 4–6 hrs 40 min 3–5 hrs
Spec formatting + QC 4–6 hrs 5 min 4–6 hrs
Material submittal 2 hrs 5 min ~2 hrs
Contract review 2–3 hrs 20 min ~2.5 hrs
Utility research 3–4 hrs 30 min 2.5–3.5 hrs
Plan QC (per sheet) 4–6 hrs 10 min ~5 hrs
Site feasibility report 20–30 hrs 3–4 hrs 16–26 hrs

Across all 15 agents: 10–15 hours freed per engineer per week. At $150/hr, that’s $1,500–2,250 in billable capacity recovered — not by cutting staff, by clearing backlog.

Under the hood

This isn’t a chatbot wrapper. It’s mostly engineered code, with model calls only where judgment is needed — seven layers running together on every job.

Supervisor

Monitors the entire fleet. Handles failures, restarts, and escalations without human intervention.

Orchestrator

Receives every incoming request, routes it to the right agent, and manages the job queue.

Specialist agents

Each built for one workflow — its own prompt stack, output format, and QC criteria. None are generic.

Skills layer

Shared knowledge every agent draws on: firm config, standards, jurisdictions, templates, project history.

Harness

The execution and testing framework. Validates agent behavior before anything touches a real job.

QC gate

Multi-agent QC gate checks every output before it leaves the system — format, completeness, and accuracy.

Every request runs through all seven layers. Every deliverable clears the QC gate. Nothing leaves until the system says it’s ready.

No new software. No UI. No training overhead.

Your team emails the AI. The AI does the work. A finished deliverable comes back. This is exactly how a fire flow request runs through Calichi’s system today — from staff email to stamped-and-ready deliverable in under four hours.

The licensed engineer is still in charge of every output.

Every deliverable is PE-reviewed and stamped

The licensed engineer reads the delivery note, checks flagged items, and stamps only after independent verification. AI deliverables are PE-ready, not stamp-ready. The final 5–10% is the licensed engineer.

Engineers make all judgment calls

Scope, risk, edge cases, design trade-offs — the engineer decides. The AI surfaces inputs, methodology, and assumptions. The engineer interprets and decides.

Client relationships are entirely human

The AI doesn’t talk to clients. It produces deliverables your team reviews and delivers. Your client relationships are unchanged.

The AI does pattern work; engineers do work that needs them

Engineers write the request (10%), the AI executes it (80%), engineer reviews and stamps (10%). Judgment, design, and problem framing stay with your team.

“AI does not replace humans — it supplements them. Employees now focus on framing problems and refining solutions over the mechanics of the solutions.”— Reco Prianto, P.E. · from the build

6 months to build it yourself. Weeks with Firma AI.

Calichi’s deployment took 6 months and 600+ hours because it was built from scratch. Firma AI deploys that proven architecture inside your firm in weeks — configured to your identity, standards, jurisdictions, and workflows. The architecture is solved. The customization is what we do for you.