What changed
FACT (source: DeepMind blog): a low-cost, low-latency model tier (Gemini 3.5 Flash) now offers agentic computer-use/vision economically viable for solo builders. FACT (source: OfficeCLI GitHub): a single headless binary can generate Office documents server-side with no Microsoft license. HYPOTHESIS: combined, these make on-site photo-to-document assembly of a FEMA-format Project Worksheet technically feasible for a solo builder.
Why now
Vision-capable agents cheap enough to run per-photo at field scale only arrived recently, and license-free server-side document generation removes the Office dependency. The claimed multi-billion-dollar Michigan/California facility-repair PA awards (INPUT ids -31,-33) are HYPOTHESIS — no citation URL was provided for the award amounts, so the specific dollar figures are unverified and must be confirmed against FEMA/state PA dashboards before use as demand evidence.
Converging signals
Cheap agentic vision (ai) + license-free document generation (dev) + a federally standardized filing burden (govmandate). The 2 CFR 200 procurement rule, the class of subrecipient filers, and the state PA portal meet at one point — a forced-filer convergence.
Customer pain
FACT (from the compliance regime, not a cited complaint): FEMA permanent-work PWs are documentation-heavy and years-later deobligations/clawbacks concentrate on procurement (2 CFR 200) and cost-reasonableness findings. Assembling a defensible site inspection + cost estimate historically takes engineering time. INFERENCE: crews and grant consultants bear this cost. No complaint URLs were supplied, so pain intensity is inferred, not evidenced.
Who pays
City/county public-works departments and the engineering/grant-management consultants who file facility-repair PWs against state PA awards. INFERENCE: consultants are the more reachable, higher-willingness buyer (white-label the tool into their workflow) than direct government procurement.
Solved today
Engineers and grant consultants manually assemble site inspections, scopes, and cost estimates (often RS Means-based), and firms bill a percentage of the award or hourly. FEMA's Grants Portal and state systems are the filing endpoint.
Why current solutions are bad
Slow, expensive (weeks of engineering time / percentage fees), and the manual procurement trail is the single most common deobligation cause — so errors surface as clawbacks years later.
Proposed product
A mobile field-capture app + backend: guided photo/dimension/pre-disaster-condition capture with GPS geotagging; vision model drafts the site inspection report and a unit-cost cost estimate; OfficeCLI renders the PW-format documents and closeout packet; a 2 CFR 200 procurement checklist validates the repair contract trail. Sold as a white-label tool to PA consultants and directly to county public-works grant staff.
MVP version
One pilot state's PW template + a single facility-repair category. Guided capture flow, geotagged photo set, auto-drafted site inspection narrative, a cost estimate seeded with a licensed unit-cost dataset, and a procurement checklist that flags missing competition/sole-source justification. Human-in-the-loop review before export — never auto-submit an unreviewed estimate.
30-day build
Confirm the actual MI/CA PA award figures and filer counts (verify INPUT ids -31/-33 against FEMA PA dashboards). Interview 5-8 PA grant consultants and 2-3 county public-works grant managers to validate pain and pricing. Secure a unit-cost data source (license RS Means-style data or a defensible alternative) — this is the critical dependency. Build the capture-to-narrative prototype.
60-day build
Ship the PW-document generation via OfficeCLI, the cost-estimate builder, and the 2 CFR 200 checker for one state template. Run the KILL TEST: field-capture a real damaged facility and have a consultant compare the auto-generated inspection + estimate against their own for cost-reasonableness and audit-survivability.
90-day revenue plan
Land 2-3 paying consultant/county pilots at a per-PW or per-seat price. Iterate on estimate defensibility from pilot feedback. Given the data-licensing and estimate-validation work, first revenue is realistically 90-180 days, not 30 — acceptable for this funded founder.
Distribution path
Direct outreach to PA/emergency-management grant consultants and county public-works grant offices; presence at APA/FEMA PA and emergency-management channels; white-label partnership with an existing consultancy as the fastest wedge.
Pricing hypothesis
Per-PW filing fee (e.g., $150-500/PW, undercutting percentage-of-award consulting) or a per-seat/per-project SaaS seat for consultants ($200-500/mo). Founder's proven FMCSA per-upload model maps directly.
Technical difficulty
Moderate-to-high: guided capture UX, reliable vision extraction of dimensions/conditions, and — hardest — a cost estimate that survives cost-reasonableness review. The estimate defensibility, not the plumbing, is the real difficulty.
Legal / regulatory risk
Moderate: producing cost estimates and procurement guidance that feed federal grant filings carries accuracy/liability exposure. Mitigate with human-in-the-loop sign-off and by positioning as a preparation tool, not a certification. No professional-engineer stamp is claimed by the software.
Platform dependency
Low platform-policy risk (submits to government systems, no deplatformer). Dependency on the Gemini Flash vision tier and on a licensed unit-cost dataset are the real external dependencies.
Founder fit
Very high on shape: a regulation-driven forced-filer flow into a government portal with per-transaction monetization — the exact FMCSA ELDT pattern the founder has already shipped. His industrial/operations and public-records background fits site-inspection and procurement-trail work. Tempered by the cost-estimating-data dependency, which is outside his prior domain.
Breakout potential
High if the estimate holds up at closeout: PW formats are federally standardized, so a working tool replicates across ~50 states and every future disaster declaration — a recurring, non-saturating market.
Final recommendation
PURSUE WITH VALIDATION GATE. Strong founder-fit forced-filer shape, but two things must be verified before real spend: (1) the MI/CA PA award scale and filer counts, and (2) that an auto-generated cost estimate survives a real cost-reasonableness / 2 CFR 200 closeout review. Run the KILL TEST early; if the estimate won't hold, pivot to the pure procurement/documentation-checklist layer (lower liability, still valuable) rather than the full estimate.
Next action
Verify the FEMA PA award figures for MI/CA (INPUT ids -31/-33) against public FEMA PA dashboards, then cold-interview 5 PA grant consultants to confirm pain, current cost, and willingness to pay for a per-PW field-capture tool.