asMade Intelligence
The system that reads all your projects, and tells you where every word comes from.
Large projects produce thousands of records a week: schedules, budgets, diaries, delivery notes, defects. asMade Intelligence connects them into one understanding: where the project drifts, why, what it costs, and what is missing to know. With a source for every finding, and without touching the data.
In the management meeting, every number becomes an argument about its source.
The schedule says one thing, the diary another, the budget a third, and what was reported from the field was not always recorded. Whoever summarises needs to know what is recorded, what was reported, what was calculated and what is simply missing. That is exactly what asMade Intelligence does.
Calculated from records
Arithmetic on recorded values, with no guessing.
A person's report
What was written or reported, with no independent verification. It stays a report, in every answer.
A link the system made
A connection between two records that the system inferred, marked as such.
Missing information
What the data does not hold, and what to update so more can be said.
From the picture of the whole portfolio to a single diary line.
Every project on one screen, ordered by what is urgent.
A senior manager sees every subscribed project: the approved budget against what was proposed and not approved, the findings by severity, and the largest schedule gap. A click opens the project's Intelligence page.
- Severity: urgent, needs action, review, keep an eye, for information
- Only what was computed from the projects' own records is shown
- Hebrew and English, in its own interface
A schedule read as it is, not as we wished.
The system reads the dependency links, calculates the critical path and compares it with what is stored in the schedule. A forecast is produced only where progress is reported. Otherwise the date is shown as the plan.
- Four link types, with lag. Cycles are exposed
- An activity's own delay, a delay passed to it, and a delay that moves the finish
- Forecast states: ok, limited, no progress, unreadable
Money, without mixing approved, proposed and paid.
The approved budget is shown apart from what was proposed and not approved. Budget changes are checked against a recorded reason, and a cost floor is shown: what has already been spent plus commitments not yet drawn, with the inclusion rule written out. It is a floor, not a forecast.
- Budget changes with no recorded reason
- Change orders against budget changes
- The budget distribution by item
What happens on site, what is recorded and what is missing.
Rules that run over the project's records and group the findings into six topics: schedule, money, supply, quality, workforce and records. Every finding is marked with how it was reached.
- Overdue tasks, delivery notes not verified
- Pending quality inspections and open defects
- Attendance gaps, silent contractors, diary delays, records discipline
Ask the project. Get an answer with sources.
An AI assistant that answers from the project's own data only. Every answer says whether it was completed, cut short or cannot be answered, and its sources are verified against what was shown to the assistant in that question. Every question is recorded, including refused ones.
- Budget, budget changes, work diaries, plan revisions and the schedule
- A source trail: id, type and key for every source
- What a person reported never looks like a record
What is live today, and what is being built.
Every capability is marked with its state: LIVE is active today, NEXT is in development, LATER is planned. We do not present as live what is not yet.
A control center for the whole portfolio
Every subscribed project on one screen: findings by severity, approved against proposed budget, and the largest gap. A separate page for each project.
Critical path and dependency network
Read from the dependency links (four types, with lag), calculated and compared with what is stored in the schedule. Cycles and contradictions are exposed. Run on a real project of 637 activities.
A forecast with honest states
Ok, limited, no progress, unreadable. The plan is never presented as a forecast.
Findings in six topics
Schedule, money, supply, quality, workforce and records. Every finding with a source label.
An AI assistant with verified citations
Answers from the project's own data only, with sources checked against what was shown to the assistant.
An evidence ledger with history
Seven source types are recorded with provenance and a data class. A changed source supersedes the earlier reading instead of overwriting it, and the system knows whether the source still exists.
Budget control and a cost floor
Approved against proposed, changes with no reason, change orders against budget changes, and a cost floor with its inclusion rule written out.
Plan and drawing knowledge
Reading plans: a sheet index, legend and element register, and a comparison with what was reported on site. Built and tested locally, not yet switched on.
A focused question when an explanation is missing
When the data does not explain a change, one question to the person who knows. The answer is stored as evidence. Proven in tests, not yet in the interface.
Channels: the WhatsApp project agent
The agent is the channel and the understanding stays here. It will ask, and not hold an understanding of its own.
The project calendar
Importing working days and holidays into the dependency arithmetic, to remove the one remaining approximation.
More sources
Periodic site reports, audit logs and delivery notes as evidence. Without counting as covered what was not read.
Location intelligence
Linking evidence to a place in the building: floor, zone, room, by the quality control location tree.
Cost forecasting
From a cost floor to a forecast, with a variance engine. Planned.
ERP and accounting connectors
Money recorded outside asMade enters as evidence, under the same source rules. Planned.
Write-back, with a person's approval
The system proposes, a person approves, and only then is anything written. Today: read-only, by decision.
More engineering disciplines in plan reading
Architecture, plumbing and HVAC, once measured. Only electrical has been tested so far.
Learning and forecast calibration
A ledger of forecasts checked against what actually happened, and calibration over time. Planned.
Built for an organisation where every read must be auditable.
Isolation between companies and projects
The company, the project and the user are set on the server from the sign-in, not from the request. A tool cannot name another project. Tested in production against a neighbouring project.
Read-only, at the database
Access to asMade data is through accounts that can only read. 25 write attempts were refused by the database.
An audit row for every call
A row for every tool call, including refusals, and for every model call. Costs reconcile with the record.
Resistance to prompt injection
Instructions planted in work diaries and records were read as content and not obeyed. Also tested with input fuzzing: 446 requests and 1,559 checks.
Usage and cost limits
Every run has a ceiling on steps, time and cost, and every company a limit on runs per hour, checked before the run starts.
Per-user permissions and a per-project subscription
Permission is checked on the server on every request. A signed-out request is refused before any run or model call exists. The attendance rules work on counts only, without personal details.
Retention: a current record is kept for the life of the project, and a superseded one may be pruned after 13 months, manually and reported. For security questions and data-processing agreements, contact us.
The understanding layer above every asMade service.
asMade holds the work. asMade Intelligence reads from it and understands it.
Questions executives ask.
What is asMade Intelligence?
Is it suited to large companies?
What is live today and what is on the way?
Does it change data in the project?
How can I trust what it says?
What happens when the data is not enough?
How are the data and the model handled?
How does it relate to the WhatsApp project agent and the other services?
How do we start?
Let your next large project explain itself.
We start with one project: a schedule with a baseline, progress reporting and a team that gives about five minutes a week. We switch it on, accompany it and show what the system understands.