GF
Giles Flythe Engineers
Reserve Study Platform · Architectural Decisions
v2 · June 2026
2
For GFE Leadership Review

Phase 1 shipped.
Phase 2 begins.

The narrative generator, the email pipeline, and the versioned QA skill are live — already giving engineers an optional head start on every draft. Phase 2 picks up where the May Discovery Call with Tyler left off — the Smart Intake direction (with an early prototype already in hand) and the FPA improvement opportunities. The June Discovery Call is the next checkpoint — where we align on how Phase 2 unfolds from here.

Audience
GFE Leadership & Engineering
Decision Class
Phase 2 · In Flight
Prepared by
Jason Cooper · wyth.ai
Phase 2 · Heading Into June

Pick up where the May call left off.

Phase 1 shipped the engine. Phase 2 listens. The May Discovery Call with Tyler advanced two threads — Smart Intake (an early prototype is already in hand) and the FPA improvement opportunities. The June Discovery Call is where we align on how Phase 2 plays out.

01
Validate

Pilot the email pipeline.

Three engineers, five real submissions each. The inbox-fetch & send-dispatcher loop has run on live test submissions since May 19 — Phase 2 puts it in front of the people whose drafts actually go to clients.

Measured — time from inbox to clean draft, time engineers spend editing, where reviewers flag the voice, which sections of the existing skill bundle are carrying the most weight.
02
Identify

Find the top three FPA quality-of-life fixes.

The May Discovery Call surfaced specific FPA improvement opportunities. Phase 2 turns those into a prioritized list — sitting with engineers and learning which three would most change how the work feels. The duplicate-and-zero-out flow came up in the May call as a leading candidate; the final list comes from the people in the tool.

Output — a prioritized, scoped list of three fixes, each described in terms of the engineer pain it removes. Ready to ship in the next wave if the June call points that direction.
03
Explore

Prototype the Smart Intake experience.

Push the Smart Intake prototype further. Sharpen the field types, the inline validation, the condition-rating UI, the moment-by-moment flow. Put it in front of engineers and learn what lands before committing to a full build.

Output — a touchable prototype and the structured-capture lessons it surfaces. The thing that lets the June call choose between building the form and extending the FPA UI — with evidence in hand, not assumptions.
The Smart Intake Prototype

Already in hand.

Phase 2 starts from a working prototype, not a blank page. Below is a walkthrough of the conversational, structured-capture flow — condition ratings, component sub-attributes, sub-area assignments. Phase 2 sharpens it, pressure-tests it with engineers, and learns from it as Phase 2 unfolds.

Built · May 2026
GFE Reserve Study Intake — Walkthrough
01
Commitment One

Engineer judgment goes in once, structured.

When you rate a roof or call a paving year, that judgment is captured the moment you make it — in structured fields the system can read directly. No re-keying from Excel into the funding plan into a Word draft.

Smart Intake takes over the engineer-facing input. The Phase 2 FPA UI patches use the same field names, the same condition-rating vocabulary, the same shape — so engineers feel the relief now and the eventual transition to the form is a UI swap, not a relearn.

Excel Workbook RETIRED Smart Intake Form Asphalt Paving Good Roofing Fair Pool · 2 Good Clubhouse HVAC Review Mailbox Stations Good Retaining Walls Good ENGINEER · STRUCTURED CAPTURE
02
Commitment Two

Your attention goes where it’s needed.

When the draft hits something that calls for your judgment — an unmatched cost, a component without a clear analog, an interpretation that’s yours to make — the system makes that hand-off explicit. You don’t hunt for what to look at.

How that hand-off shows up is a conversation, not a decision: highlighted inline in the draft, listed in an insights email, queued as a review note. The commitment is the principle — the system never silently fills what it isn’t sure about.

! ! !
03
Commitment Three

Reports are built fresh, every time.

Instead of starting from a copy of a previous community’s report and editing it down, every report composes from your inputs and a clean renderer. Less copy-paste. Less drift. None of the carryovers that happen when a draft inherits material from an earlier study.

The transition is staged. A template-lint pass catches the specific patterns we know about today. A programmatic renderer takes over once the QA data shows the template file itself is the limiting factor — not the inputs flowing into it.

Report Template CARRYOVER BUGS def render(): doc.add(...) doc.add(...) return doc code { community, components, financials } JSON Clean .docx PROGRAMMATIC
The Lineage

What stays. What extends. What's replaced. What's new.

Every component of GFE's current workflow is accounted for. Most of what works keeps working; the parts that introduce drift or carryover get retired; a small set of new capabilities is added. Nothing is silently dropped.

Preserve
Extend
Replace
Add New
Funding Plan App · backend
Cash-flow projections, threshold analysis, alternative modeling. The computation engine stays — it's the financial source of truth.
GF Voice & Vocabulary
Curated patterns carry forward as prompt inputs.
Historical Exemplars
The firm's best past reports join the knowledge base as voice anchors.
Email Pipeline
Inbox fetch + send dispatcher remain.
Narrative Generation
Existing logic extended into a versioned, retrieval-fed engine.
Lessons Learned Corpus
Becomes the input to an approval-gated learning loop.
QA Pass
Promoted from skill checklist to deterministic test suite.
Past Report Library
Indexed for vector retrieval and starting-template match.
Excel as Engineer Input
Retired once Smart Intake is validated and adopted. Until then, the email-attached workbook remains the engineer's input channel.
Report Template.docx
Path to a programmatic renderer, gated on QA data. Template-lint pass first; full renderer only when the data says the template file is the limiting factor.
Funding Plan App · UI
Phase 2: identify the top three FPA UI improvements engineers most want, and prototype Smart Intake in parallel. Longer term: Smart Intake form writes to the FPA backend via API and replaces the engineer-facing screens.
Skill-Bundle Generator
Path to a code-first generator (shadow mode during cutover); skill bundle continues delivering production drafts during Phase 2.
Smart Intake Form
Wizard for structured capture of engineer judgment.
ReserveStudy JSON Schema
Versioned data contract spanning intake → generator → platform.
Starting-Template Match
Pre-fill from the closest past community by component profile.
Cost-Variance Flags
Inline reconciliation against the internal cost database.
Self-Learning Loop
Diff extraction from reviewer corrections, gated by approval.
The June Discovery Call

One decision unblocks the rest.

The June Discovery Call is where Phase 2 gets its direction. Pilot data from real submissions, the top three FPA improvements engineers most want, and the evolved Smart Intake prototype — that's the working surface. The call's job is the sequencing decision below. The answer determines what we build first.

June Discovery Call
Sequencing · One decision

Do we build the Smart Intake form next — or keep iterating the FPA UI with Smart-Intake-aligned patches and defer the full form build?

Path A

Smart Intake form — first.

Commit to building the form once. Engineers transition off the FPA UI for engineer-facing capture. Higher ceiling on engineer experience; larger upfront build; one mental-model migration. Justified if Phase 2 data shows the FPA UI is the limiting factor on quality, not just on time.

Path B

FPA UI patches — keep going.

Take on a wave of FPA UI patches as the next workstream — starting with the three Phase 2 surfaces. Faster relief; lower risk; defers the form build. Justified if Phase 2 shows targeted FPA fixes alone would change the day-to-day enough that the full form isn’t urgent.