Product roadmap — trade depth, intelligence & packaging

Product context: Estym8 is built from scratch as an AI-first construction preconstruction platform—not a legacy takeoff stack with AI bolted on. AI runs across the product: bid-package ingestion and classification, multi-model takeoff and vision, plan intelligence, cross-file synthesis, Estee, estimate-to-submittal draft review, and optimization recommendations. Canonical framing: AI-first positioning.

Served at: estym8.ai/docs/roadmap (slug roadmap in lib/docs/public-docs.ts). The filename ROADMAP_BEYOND_MEP.md is unchanged so existing repo links keep working.

What this doc is: Forward-looking depth, new vertical outputs, and packaging—stacked on a shipped whole-project surface. It is not where we describe “coming soon: folder upload”; bid-package ingest (classify → plan → execute), verbatim harvest, code-edition mismatch alerts, cross-file intelligence, and the multi-discipline narrative report are live—see folder workflow and investor overview.

“Beyond MEP” in one sentence: MEP remains the deepest automation and calibration lane today; this roadmap tracks depth and packaging on verticals that already have structured schemas shipped (doors/windows/steel/rebar/utilities/demo/consistency — see UNIFIED §P3) plus MEP extensions (air balance, HVAC recommendations, …) and estimate-to-submittal polish — not a second product.

Related references: Takeoff definitions + tiers → takeoff verticals; execution queue → repo UNIFIED_BACKLOG.md (Now / Next / Later); this page is public narrative, not the work queue.

Baseline already shipped (do not re-sell as roadmap)

Theme	Status
Folder / bid-package upload (multi-type), analyze → plan → document → execute	Shipped
6+ disciplines on one pipeline (E/M/P + civil / arch / structural); multi-file bundles	Shipped
Per-document “As Printed on the Sheets” harvest	Shipped
Code-edition mismatch rollup on the project	Shipped
Post-folder cross-file intelligence (consistency, coordination, RFI seeds)	Shipped
Multi-discipline narrative report (Markdown / DOCX / PDF) + bid-bucket reconciliation	Shipped
Single-PDF path + optional symbols-only takeoff → promote to estimate	Shipped
L1 structured takeoff verticals (doors, windows, steel, rebar, horizontal utilities, demo/temps, discipline consistency)	Shipped (depth/calibration ongoing)
Estimate-to-submittal (catalog, packages, register, compliance matrix, merged PDF, draft spec AI)	GA-track, evolving
Portfolio discovery SKUs (ClearToPermit / ClearToIssue)	Discovery — workflow apps with AI assist (not chatbots); staff keep authority; see product briefs

When do the sections below land?

L1 verticals already have schemas (doors/windows/steel/rebar/utilities/demo/consistency). Roadmap themes below are depth, calibration, and new L2/L3 slices (concrete, drywall, air balance, …)—not “first-time invent the vertical.” Effort is prompt/schema/UI/export + real-plan validation.
MEP extensions (air balance, HVAC recommendations, insulation notes): extensions to the current estimate model—see takeoff verticals. Queue in UNIFIED Later until promoted to Now.
Progress / as-built: Own epic when pulled into the backlog—not implied by folder GA.
Accessibility panel: Scope when public-sector / ADA customers require it.
ClearToPermit / ClearToIssue / Revit: Suite discovery — Peoria-shaped municipal path may elevate; Revit import flagship for ClearToPermit; UX = workflow not chatbot — municipal brief · paul-notes; not Estym8 takeoff thrash.

Summary: Folder GA and L1 structured verticals are shipped. Public roadmap below = forward depth + packaging. Execution order lives only in UNIFIED_BACKLOG.md.

Roadmap themes (prioritized by feedback; execution in backlog)

1. Air balance + HVAC recommendations

Air balance: From mechanical plans and equipment schedules, derive airflow (CFM) by zone/duct/diffuser and output balancing-ready report or pre-filled balancing form.
HVAC recommendations: Equipment sizing, system type, or efficiency recommendations where schedules/specs/locale support them.
Note: Stakeholder feedback often bundles these; include both when scoping.

2. Concrete / earthwork (when the project supports it)

Slab, footings, stem walls, curb/gutter from floor plans and sections → quantities (CY) and concrete takeoff.
Cut/fill from site/grading when sheets contain usable geometry/context.
Conditional: Surface when the drawing set clearly includes concrete/sitework (sheet list, scope, classifier hints)—avoid empty promises on sets with no civil/structural signal.

2b. Pool construction / pool builders (site-forward)

Context: Pool work is often in-ground / site-forward: layout, circulation, equipment, limited power—different sheet mix than stacked commercial MEP.
Today: Same PDF → takeoff pipeline applies when legends, schedules, and countable scope map to current extraction; quality follows sheet fidelity and MEP-forward calibration.
Future slice: With bid-package folder already standard, “pool-adjacent” depth is a scoped vertical (equipment counts, tie-ins to concrete/earthwork §2 when sheets support it)—validate on real pool plan sets before marketing a dedicated SKU.

3. Drywall / partitions / finishes

Partition types and linear feet by type/height; optional room-by-room SF for finishes.

4. Doors, frames, hardware + windows

Door schedule + plans → counts by type/size, frames, hardware, opening dimensions; flag missing refs.
Windows: Window schedule + plans → counts, rough openings, key details.

5. Insulation / envelope (when it applies)

From envelope, roof, or wall sections, recommend insulation type/R-value when visible or typical for project type/locale.
Conditional: Only when plans or context support defensible output.

6. Roofing and waterproofing

Roof area by system, perimeter/flashing length, penetration counts.

7. Structural (rebar, steel)

Rebar from foundation/slab details; steel members/sizes/connections for first-pass structural takeoff.

8. Cross-discipline consistency (deeper than today)

Cross-check room names/numbers, areas, equipment counts, dimensions across arch/MEP/schedules.
Today: Harvest + cross-file intelligence + narrative report already surface many conflicts; this theme is deeper automation and reviewer UX, not “from zero.”

9. Progress and as-built

Same PDF pipeline for planned vs installed by area/zone when revised PDFs or markups exist; optional manual “installed count” inputs.
Requires workflow + storage design before UI promises.

10. Model independence — proprietary domain models (post-funding, data-gated)

Today: orchestrate best-in-class third-party models (Claude and others) behind a model-agnostic pipeline — frontier capability at inference-only cost, no training spend. The model layer is swappable by stage (text / vision / structured).
The asset that unlocks this: the proprietary ground-truth + calibration corpus (real won projects + human takeoffs) and the regression harness. Model independence is gated on corpus depth, not a day-one effort.
Theme: train fine-tuned / distilled domain models for the narrow, repeatable tasks (symbol/device classification, schedule extraction, count reconciliation) and slot them in task-by-task wherever they beat third-party calls on regression. Payoffs: lower inference cost, reduced vendor dependence/lock-in, domain accuracy.
Explicitly out of scope: building a general-purpose foundation model from scratch — wrong cost class and not the moat. Sequencing follows corpus growth and unit-economics signals, not a fixed date.

Accessibility control

Stakeholder ask: Optional accessibility panel (saturation, contrast, font sizing, spacing, cursor, blink blocking, captions/magnifier, statement + feedback).
Implementation: Offer when contract or sector requires it (public sector, ADA-driven); per-project or per-org toggle so default UI stays clean.

Already implemented

Elevation and survey in demographics: Master prompt + schema expose demographicData.elevationSummary and demographicData.surveyInfo; Demographics UI and plan metadata surfaces them when present.

Schema / prompt changes (elevation/survey — done)

DemographicDataSchema: optional elevationSummary, surveyInfo.
Master prompts: locale & demographics section writes into demographicData; JSON schema includes fields.
Estimate / plan metadata UI: hasDemographics and display paths updated.