Ingest
Connect the scattered records
Upload P6 XER schedules, six-week lookaheads, daily reports, risk logs, and full specifications. tomok keeps every file as reviewable evidence.
Construction intelligence for heavy civil
See what puts delivery at risk.
A delayed task pulled 15 people into a 3-hour meeting to ask a single question. We built an AI that answers it in three minutes.
tomok connects P6 schedules, lookaheads, daily reports, and thousand-page specs into one project model — every answer traceable to its source.
- $2B
- tunnel program managed
- 1,200
- spec pages parsed
- ~3 min
- to answer “why late?”
Watch the demo
Three minutes, start to answer.
The full walkthrough on real B&P Tunnel data — from uploading the records to tracing a delay back to its source.
Prefer to read the code? View the source on GitHub
From the field
Hear the problem from someone who lives it.
Jae has spent 16 years in heavy civil and managed $15 billion worth of projects. Here he explains why project data stays siloed across cost, time, and field systems — and what that costs a job every single day.
tomok is built with and for practitioners like Jae, on real records from a live $2B program.
See it in action
One project view. Every answer traceable.
tomok ranks schedule movement, production gaps, document exposure, and open actions — then links each signal to the record behind it.
B&P Tunnel Program · $2.0B
B&P Tunnel Program
Owner delivery overview
QW
Delivery confidence
At risk
Forecast finish
+18 days
Sources ready
4 / 4
Progress outlook
Planned vs. actual through 01 Aug
PlannedActual
W1
W2
W3
W4
W5
W6
Needs attention
Ranked by delivery impact
Line N support excavation
7d late
Subcontract award package
Due Fri
Daily report coverage
Current
Every signal opens its source
How it works
Deterministic where it must be. AI where it helps.
Sending a whole P6 file to a model invites hallucination. tomok separates what must be mathematically exact from what benefits from reasoning.
Normalize
Parse the schedule into real tables
Rather than send a whole XER to a model, tomok parses it into normalized PostgreSQL tables — activities, dates, total float, constraints, and relationships.
Reason
Retrieve only what matters
Purpose-built tools pull the exact critical and near-critical activities, then compare them against the specific daily reports and lookaheads that explain movement.
Act
Put the signal in front of the right role
Rank what threatens delivery, attach a recommendation, owner, and due date — and link each finding straight back to the activity and document behind it.
The problem, from the field
Both sides of the job lose time to the same gap.
On the real Amtrak B&P Tunnel program, data is siloed across cost, time, and field systems — so the truth takes a room full of people to assemble.
“Pulled 15 people into a 3-hour meeting to answer one question: why is this task behind the baseline? It cost $10,000 in billable time.”
“Reads a 1,200-page spec to build the submittal register by hand — days of work — while a single day of schedule delay runs $500,000 in fees.”
What it does
Six workflows that used to take weeks.
A focused operating surface for owners, project managers, schedulers, field teams, and document control.
Answer why a task is late
Ask in plain language. tomok reconciles baseline, latest update, lookaheads, and dailies, then explains the delay with the exact P6 activities that drove it.
Detect risk from daily reports
When the field uploads a daily report, tomok re-runs the forecast overnight and surfaces emerging schedule, production, and procurement risk by morning.
Extract the submittal register
Turn a 1,200-page specification into a reviewable submittal list, split by CSI division, with exact section references — generic TOC claims rejected.
Draft baseline activities
Convert extracted requirements into draft preparation, review, procurement, and delivery activities — exportable natively back to XER for a scheduler.
Read P6 without the license
Open Primavera XER files directly — baseline, WBS, float, and relationships — so all 60 people on the job can read the plan, not just the six schedulers.
Trace every finding to its source
No unexplained scores. Each conclusion opens the exact schedule activity, report, or specification passage — so teams verify instead of trusting.
How it's built
A hybrid of exact logic and grounded reasoning.
Built in days with Next.js, FastAPI, and the OpenAI Agents SDK, ChatKit, File Search, and the Responses API — on a foundation designed for trust on sensitive construction data.
The model is not the product. Reliable construction intelligence needs domain-specific tools, structured data, and strict permissions.
Deterministic project-control core
Float, date variance, and schedule comparisons computed exactly — never guessed by a model.
Normalized schedule tables
XER and Primavera XML parsed into PostgreSQL so tools retrieve precise activities and dates.
Bounded specification extraction
Large specs processed in CSI-division groups into structured, source-referenced submittal records.
Default-deny confidential data
Resource, labor, equipment, and cost records stay hidden unless membership explicitly permits them.
What the agent produces
Every output keeps a path back to its evidence.
Delay analysis
Why a task slipped, ranked by delivery impact, with the driving activities and float.
Source-linkedEarly-warning forecast
Next-morning schedule, production, and procurement warnings after each daily report.
Source-linkedSubmittal register
Source-referenced submittals organized by location and CSI division, ready to review.
Source-linkedWhere it stands
Already managing a $2 billion tunnel program.
- 100K+
- P6 activities parsed into tables
- 3 hrs → 3 min
- to answer a delay question
- $4K
- P6 viewer seat, replaced
- 60
- team members who can read the schedule
The AI operating system for heavy civil.
Put the next project decision on firmer ground — with the exact activity and document behind every answer.