NowIn private development with real DOT plan sets
Engineering drawings,
understood.
Civil projects hold enormous amounts of information trapped inside 2D plan sets. Mathyra is building the intelligence layer that understands it — geometry, relationships, and validation, with a source for every value.
Illustration — a mock plan lifted into a mock corridor. Not engine output.
The problem
A plan set is a database that only humans can read.
Every sheet states facts: stations, elevations, curve data, control coordinates, pipe sizes, side slopes. The facts reference each other across hundreds of pages — and the moment the PDF is issued, all of that structure is flattened into ink. Software today mostly shovels the ink. The structure is the value, and the structure is what gets lost.
The transformation
From ink to model, in one continuous shot.
A 2D plan set becoming 3D, machine-readable engineering data — geometry lifting off the sheet, values attaching, sections rising, a surface lofting — without ever losing the thread back to the drawing. Scroll, and watch it run.
A contract plan set, exactly as issued. Ink, labels, conventions — everything the drafter meant, flattened onto paper.
The linework lifts off the page as real coordinates. Nothing is invented: the 3D curve is built from the same points the drawing states.
Stations, elevations, offsets — read as machine-readable engineering values and pinned to the geometry they describe.
Every cross-section template stands up at its own station, carrying the shape the sheet drew for it.
Sections loft into a corridor; terrain gathers around it. Built only from geometry that survived validation.
A machine-readable engineering model — and the thread still runs: pick any vertex and it walks back to the sheet it came from.
What Mathyra understands
One drawing. Six kinds of evidence.
Understanding a sheet means reading all of them together — and knowing which one each conclusion rests on.
The differentiator
When the evidence is thin,
Mathyra says so.
Extraction tools produce data. The question that matters is whether the data can be trusted — and trust starts with a system that refuses to guess.
When a value can't be established from the ink — a label too ambiguous, a datum that can't be corroborated, two readings that disagree — Mathyra's engine flags it for review instead of silently picking one. Every refusal is counted, named, and carries the evidence that triggered it. This is how the engine works today, not a roadmap item.
- Elevation
- 435.20 ft
- Confidence
- Insufficient evidence
⚠ REVIEW REQUIRED — not exported
Cross-sheet intelligence In development
The same station lives on four sheets.
Mathyra compares the evidence.
Plan, profile, cross-section, control — four drawings of one physical place. Because Mathyra reads them as one model, it can ask a question no single sheet can answer: do they agree?
E 2,101,554.02
EL 431.884 · NAVD 88
COMB. FACTOR 0.99993
✓ EVIDENCE AGREES — profile and section state the same elevation at 142+00.
Illustration with mock values. Cross-sheet reconciliation is running in development today — the engine already scores real plan-profile pairs and refuses the joins it cannot prove.
Provenance Working today
Every result has a source.
Pick any vertex of the model and Mathyra can walk you back to the exact ink that produced it — the sheet, the station, the location on the page. In the engine today, every exported coordinate carries this trace; a value that cannot name its source does not ship.
drag to orbit · click a highlighted vertex
From drawing to model
The road from ink to engineering.
PDF Working today
A contract plan set, exactly as issued — vector ink, text, and everything the drafter meant by it.
Geometry Working today
Linework recovered as geometry with real coordinates — calibrated against the sheet’s own rulers and scale statements, deterministically.
Structured data Working today
Stations, offsets, elevations and their relationships, exported as machine-readable engineering data with per-value provenance.
Validated objects In development
Cross-sheet reconciliation, alignment attribution, datum verification — every value tested against the rest of the plan set before it is believed.
Spatial model The vision
Surfaces, corridors and alignments assembled from validated geometry only — a 3D model whose every vertex can cite its sheet.
Engineering workflows The vision
Takeoff, machine control, design QA, as-built comparison — downstream work running on evidence instead of re-keyed data.
How it works
Not “upload PDF, ask the model.”
A language model alone will happily guess. Mathyra is a hybrid system: AI where perception is genuinely hard, deterministic geometry and engineering rules everywhere else — so the same plan set produces the same model every time, and every claim is checked before it ships.
AI PERCEPTION
Reads the ambiguous parts — degraded text, drafting conventions, context a rule can’t enumerate.
GEOMETRY ENGINE
Deterministic computation of coordinates, scales and spatial relationships. Same input, same output, byte for byte.
ENGINEERING RULES
Checks the physics of the document: monotonic stations, plausible grades, datums that corroborate.
VALIDATION
Tests every result against the rest of the plan set — and refuses the ones that fail, visibly.
PROVENANCE
Records where each value came from, so the result is an argument you can audit, not an answer you must trust.
The category
Most tools extract something.
Mathyra is built to understand.
Takeoff software, PDF extractors, AI document tools, CAD converters — each pulls something out of a drawing, and each serves real needs. What they generally share is the shape of the output: data without the relationships, and results without the evidence.
Mathyra's ambition is different: read the plan set the way a engineer does — as one interconnected model — then hand you structured data that can prove where it came from and show where it disagrees with itself.
- Geometry, not pixels — coordinates calibrated by the sheet's own rulers.
- Relationships, not rows — stations joined across plan, profile and section.
- Refusal, not guessing — thin evidence becomes a flag, never a silent value.
- Provenance, not trust — every value cites its sheet and its ink.
Early access
Be there when the drawings wake up.
Mathyra is in private development with real DOT plan sets. Join the list and tell us what you work with — early access will be shaped by the people on it.