Khatam — Markup intelligence

Khatam — Markup intelligence

Canonical phrase suggestions

The sparkle button next to a markup description ranks your draft against a curated AEC phrase corpus and proposes the top alternatives. Lifts cost-engine matching accuracy because the engine sees consistent vocabulary. See Markup Toolbar and Cost Impact.

Semantic markup search

The markup sidebar's search ranks results by meaning, not just substring match. Searching "gypsum" finds markups about drywall, wallboard, and sheetrock even when those exact words weren't typed.

Similar-markup detection on save

When you add a new markup whose description closely matches an existing one (different wording, same intent), Khatam surfaces a non-blocking purple toast. The toast is a suggestion; your new markup is created regardless.

Markup insights

When you select a markup whose description is long enough to be informative, Khatam surfaces a compact row in the sidebar with up to four kinds of suggestion: a likely action (remove, add, modify, relocate, substitute, no-change), materials / locations / elements / dimensions / prices / scales it found in the text, a category for the document the markup most likely belongs to, and any dates worth flagging (deadlines, durations, historical references). Every suggestion needs a click to take effect; dismissing one is free. The full breakdown is described below.

Organize layers

The markup sidebar's Organize layers action reads each markup's discipline, cost, and intent signals and proposes named layers (Cost Items, RFIs, Demo, and per-discipline buckets), then creates them and moves the markups in one click. It reuses an existing same-named layer and merges duplicates, so running it again is safe. The Layers toolbar glows briefly to show you what changed.

Style audit

The sidebar's Style audit flags the markups missing something most of their peers have (a subject, units, a cost code), so an inconsistent set is easy to spot before you share it. Click an outlier to jump straight to it.

Conversational filter

Type a plain-language filter in the markup sidebar (for example, "pending clouds on Level 2") and Khatam maps it onto the tool, status, layer, and author filters, routing whatever is left over to semantic search.

Markup to RFI draft

Turn a markup into an RFI in one click. Khatam pre-fills the issue title, a composed description, and the RFI category from the markup's text. You review and edit before creating; nothing is filed automatically.

What the markup insights row shows you

Khatam reads the description on the markup you've selected and shows a single collapsed line below the metadata row. The line reads Khatam (N) ▶, where N is the total number of suggestions across all four families. Clicking the line expands the row; clicking again collapses it. If Khatam had nothing useful to say, the line doesn't appear at all, with no empty chrome, no wasted vertical space.

Expanding the row reveals up to four suggestion families, each in its own strip:

• Action: a single likely verb for what the markup is asking for, with a confidence number in parentheses. One Confirm button to accept; one Dismiss button to reject.
• Detected items: a wrappable row of chips, one per item Khatam found in the description. Each chip carries a type tag (material, location, element, dimension, price, scale, etc.), the chunk of text it covers, and a small ✕ to reject that one chip without touching the rest. Confirm all and Dismiss all buttons sit on their own row beneath the chips.
• Category: Khatam's guess at what kind of drawing the markup most likely belongs to (architectural, structural, mechanical, electrical, plumbing, civil, or specification). Same Confirm / Dismiss shape as Action.
• Dates: a chip row like Detected items, but for dates. Tags include deadline, duration, historical, event- relative, and recurring. The chip text is the date phrase Khatam found in the description verbatim. No chip click here ever creates a calendar event, schedule milestone, or task due-date. See the closing disclaimer below.

Khatam only shows a suggestion when its own confidence clears a per-family floor: 0.70 for the single-pick families (Action, Category), 0.60 for the chip families (Detected items, Dates). Below those floors the suggestion is suppressed entirely, so a one-word description like "gypsum" stays clean instead of being decorated with low-confidence guesses.