Product 6 min read

Linarc's Lia Hunts Submittal Requirements Outside the Section Labelled 'Submittals'

Announced 17 November and described as live in production: four document-reading capabilities inside Linarc's construction management platform — plan set indexing, drawing revision comparison, submittal register generation and contract term extraction. Every item Lia produces carries a confidence level and a citation back to the source document, and a person approves it before it is shared.

A hand-drafted floor plan sheet titled Second Level, scale one-eighth inch to a foot, lying under a mechanical pencil, a metal scale rule, an eraser, a roll of tape and a case of drafting pens.

Linarc Inc. announced Lia on 17 November from a Los Angeles dateline — an AI assistant inside its construction management platform, aimed at the documents that define a project before anyone builds anything. The release went out over GlobeNewswire, which blocks automated fetches; this publication read it through the syndicated copy carried by The Manila Times, which carries the full text under a GlobeNewswire credit.

The framing is familiar. “Every construction project begins with thousands of details buried in drawings, specifications and contracts,” the release says, and turning them into usable project information “has traditionally required hours of manual review, data entry and cross-referencing.” What distinguishes the announcement from the general run of document-AI releases is that it names four specific capabilities, describes Lia as “now live in production” and the four as its “production capabilities,” and attaches a disclosure requirement to every output.

Four Capabilities, Named Narrowly

Intelligent Plan Set Extraction. Visual language models read the drawing sheets and build a structured sheet index, “capturing sheet names, codes, revisions, dimensions and scales — even when a single sheet contains multiple drawings.” That last clause is the tell of someone who has actually looked at a plan set. A sheet carrying several details at several scales is common, and it is where sheet indexing that assumes one drawing per sheet falls over.

Drawing Revision Intelligence. Lia compares sheet versions to identify added or removed drawing elements, annotation changes and title block revisions — “including changes that were never clouded or documented in the revision block.”

That is the sentence worth stopping on. Revision clouds and revision-block entries are how a design team declares what changed. A change carrying neither is, by construction, the one nobody downstream was told about, and that is how a superseded detail reaches the field. Finding it is work a diligent human reviewer does badly, not for want of skill but because the comparison means reading hundreds of sheets against their previous versions, mark by mark.

Linarc then connects the finding to the commercial record: teams “can link identified differences to existing RFIs or initiate new RFIs and change orders directly from the sheet, with marked-up comparisons attached.” The output is not a report. It is an RFI or a change order with the evidence stapled to it.

Automated Submittal Generation. Lia analyses the specifications to build submittal registers and individual submittals, “identifying requirements throughout the specification — not just within sections explicitly labeled ‘Submittals.’” Each generated item “maintains a reference to its source section for verification and traceability.”

On this publication’s reading, this is the quiet claim with the most money behind it. A submittal obligation can be written anywhere in a spec section — in quality assurance, in delivery and handling, in a product clause that requires a manufacturer’s certificate. A register assembled only from the paragraphs headed “Submittals” is a register that can be missing items, and a missed submittal surfaces as a delayed approval at the moment the material was needed.

Contract Intelligence. Lia extracts “project milestones, liquidated damages, payment terms, retainage requirements, notice provisions and schedules of values”. Notice provisions are the quietest item on that list: a clause setting how many days a contractor has to give notice of a claim can govern whether the claim can be brought at all, and it sits in the general conditions rather than in the commercial summary at the front.

The Disclosure Requirement

The release makes the point in a bolded lead-in to its own sentence — “Built for accountability, not blind automation” — and the mechanism it introduces is two things rather than one.

Every AI-generated item “includes a confidence level and a citation linking it to the original source document.” And: “Project teams review and approve extracted information before it is shared, ensuring that contractors retain control over project decisions.”

A citation is traceability — it lets a reviewer check the extraction against the sheet or the spec paragraph it came from. A confidence level is triage — it tells the reviewer where to spend the hour they have. On this publication’s reading, shipping both is what makes the review step affordable rather than merely mandatory — which is the difference between a human-in-the-loop claim that survives a live job and one that gets switched off in week three.

Linarc’s own site states the same boundary in five words on the Lia feature: “Nothing posts without your review.”

Shanthi Rajan, founder and CEO of Linarc, is the release’s only quoted voice:

“Lia is designed to do more than answer questions or summarize documents. Lia takes on the time-consuming work of turning project information into something construction teams can actually use.”

She adds that from the first day of a project, Lia “helps teams identify requirements, uncover changes and move work forward, while keeping contractors in control of every decision. And we’re just getting started.”

What Is Shipped and What Is Not

The release is clear about the line between the two. Beyond the four production capabilities, “specification-driven procurement workflows” are “already in quality assurance” and “AI-assisted scheduling” is “in development.” The stated ambition is to extend Lia “beyond document intelligence to support planning, procurement, project management, financial controls and field execution”.

So: four things live, one in QA, one in development, and a direction of travel. The release gives no dates for the two unshipped items.

It also gives no performance figures — no extraction accuracy, no time-saved number, no named customer. This publication could not establish from the release or from Linarc’s site how the confidence levels are calibrated, what scale they use, or what thresholds, if any, gate an item from reaching a register. Those are the questions a contractor evaluating it would ask second, immediately after asking what it reads.

Linarc’s own site lists offices in Los Angeles, Austin and Chennai, and sells a modular platform — schedule, field, job cost, RFIs and submittals, pay applications — to commercial general contractors, specialty contractors, fabricators, and owners and developers. This publication found no primary source for the company’s funding history and is therefore not reporting one.

Analysis: The Pre-Construction Document Layer Is Getting Crowded

What follows is this newsroom’s reading, not a claim Linarc makes.

Lia lands in one of the most contested parts of the construction AI market. Reading plans, specs and contracts and turning them into structured obligations is where Document Crunch went when it extended its risk platform from contracts to drawings on 21 October, with source-linked citations; where Asite pushed its agents when it moved three of five of them upstream to check information as it arrives; and where qapture pointed qentin at buildings that already exist and published the sentence where it stops.

The differentiator Linarc is reaching for is not the reading. It is the write-back: an extracted requirement that becomes a row in a submittal register inside the same system that holds the schedule, the cost codes and the pay application. Document intelligence that produces a PDF summary leaves the data entry where it was. Document intelligence that produces a reviewable register entry removes it.

Whether that holds depends entirely on the quality of what reaches the register — which is not something any launch release, Linarc’s included, can settle. What can be said from the document is that the company has named its capabilities narrowly enough to be checked against a real plan set, and has put a confidence level and a citation on every item it asks a contractor to trust.