case study

Construction Project Coordination: Private AI for Permits, Procurement, and Progress

A Bangkok construction firm tracks permits, coordinates subcontractors, and manages procurement across five active projects using a private AI system.

active projects

5 residential and commercial

daily messages processed

120+ across Line and email

data sources

4 (permits, procurement, subcontractors, site reports)

the problem

Coordination across five active sites

The firm runs five active projects across Bangkok at any given time. Residential builds, commercial fit-outs, renovation contracts. Each one generates its own stream of permit applications, subcontractor communications, material orders, and daily site reports.

Before the system, coordination ran through Line group chats, email threads, shared spreadsheets, and physical paperwork. A project manager for a condominium renovation might have permit status in an email from the district office, subcontractor schedules in a Line group with 40 participants, and procurement timelines in a Google Sheet maintained by a different team.

None of it was connected. Asking "which projects have permits expiring this month" required calling three people. Asking "did the steel delivery for the Sukhumvit site actually arrive" meant scrolling through a group chat.

The information existed, but assembling it into a decision took hours every day. For a firm managing 80 employees and dozens of subcontractor relationships, the managing director spent mornings chasing updates instead of making decisions.

the build

What was built

The system runs on a server in the firm's Bangkok office. It connects to four data sources: permit records entered by the admin team, procurement tracking from purchasing, subcontractor communications from Line and email, and daily site reports from supervisors.

Permit tracking is structured. When a permit application is filed, the admin team logs it with the project, permit type, submission date, and expected response window. The system monitors aging permits, flags ones approaching deadline, and alerts the project manager when action is needed.

If a building permit has been pending for 28 days and the typical turnaround is 30, the project manager gets a message before it becomes urgent. The logic compares elapsed time against expected turnaround and notifies with enough margin to act.

Subcontractor communication runs through a monitored channel. When a subcontractor confirms a delivery date over Line, or an electrician sends an updated timeline by email, the system captures the commitment and tracks it against the project schedule. If a confirmed date passes without a corresponding site report entry, the system flags the gap.

Procurement works the same way. Material orders are logged with expected delivery dates, quantities, and the project they belong to. When prices change or lead times shift, the system surfaces the variance so the procurement team can act before it cascades into a schedule delay.

Daily site reports are submitted by supervisors through a simple form: photos, crew count, work completed, issues encountered. The system compiles these into a daily summary for each project and a consolidated view for the managing director.

The entire system is orchestrated by OpenClaw. Each data source has its own plugin, the cron scheduler generates morning briefings and weekly procurement summaries, and the messaging layer delivers through the channels the team already uses.

daily use

What it looks like day to day

The managing director's morning starts with a consolidated briefing. Five projects, one message: which sites are on schedule, which have open issues from yesterday, which permits need attention this week, which deliveries are expected today.

Follow-up questions are conversational. "What's the permit status on the Rama IX project" returns approved, pending, and upcoming submissions with dates. "Which subcontractors haven't confirmed next week's schedule" returns names and last communication dates. "Show me yesterday's site report for Thonglor" pulls the supervisor's entry with photos, crew count, and notes.

The procurement team tracks material orders and supplier confirmations across all projects in one place. A supplier emails a revised quote, and the weekly summary shows the delta alongside open orders and items at risk of delay.

Site supervisors submit daily reports in under ten minutes through a simple form. The system compiles, compares against schedule, and surfaces anything that looks off.

the result

What changed

The managing director makes decisions based on current, consolidated data instead of fragmented updates from five sources.

The Monday meeting used to be a 90-minute status update session where each project manager reported verbally from memory. With the system, the meeting starts with everyone looking at the same briefing. The conversation moves to decisions and problem-solving because the status is already known.

Permit tracking went from reactive to proactive. In the first three months, two permits that would have lapsed were renewed on time because the system flagged them ten days before expiry.

Procurement visibility reduced material delays. A concrete supplier who consistently delivers three days late is visible in the data, and that visibility drives conversations that would not have happened when delivery tracking lived in scattered Line messages.

the stack

Technical details

The core is a Node.js backend backed by SQLite. Every permit record, procurement order, subcontractor communication, and site report lives in one database on the firm's server. OpenClaw's plugin system wraps each query surface with a typed schema so the agent can answer natural language questions against the same data the automated reports use.

Data ingestion uses three paths: structured input through web forms for permits and site reports, email monitoring for supplier confirmations, and Line message ingestion for subcontractor communications. Each path is independent, and adding a new source means writing an ingestion handler without redesigning the system.

All project data stays on the firm's server. Subcontractor names, permit details, procurement pricing, and site photos never pass through external APIs. The language model handles natural language interaction and briefing generation. The underlying data storage, querying, and comparison logic is deterministic code.

OpenClaw provides the orchestration layer: plugin registration for each data source, cron scheduling for daily briefings and weekly rollups, and multi-channel delivery so the same report goes to Line, email, or a web interface depending on the recipient.

services

Private AI systems for construction, professional services, healthcare, and individuals.

how it works

Deployment model, privacy architecture, and the engagement process.

your project data belongs on your server.

The same architecture that coordinates construction projects powers health tracking, document review, and patient intake. The conversation starts with how your team actually works today.

book a consultation