Case Study · Boots On The Ground AI
Boots: an AI teammate for the people who run the business.
Boots is the small business owner’s AI teammate: upload the documents that run your business, then ask anything — by voice or text, in English or Spanish. Designed, specified, and built by one person: Damara Carlentine, founder and principal of Boots On The Ground AI, LLC, directing AI coding tools under her own specifications, plans, and review.
Multi-tenant conversational AI platform · askboots.com
Workflow design, specifications, architecture, implementation, handoff documentation
Rebranded as Boots and extended from July 2026
Row-level security, Anthropic & OpenAI models, voice & PWA, Docker
Where Boots came from: the platform began in early 2026 as a client-commissioned build, won through a competitive RFP against national agencies. When that engagement ended, Damara rebranded the codebase she had authored as Boots — and kept shipping. The client is not named on this page, by choice.
Built for work that slips through the cracks.
Frontline businesses run on shift work. In quick-service restaurants, annual hourly turnover runs 130–160%, and roughly half of new crew leave within 90 days. The knowledge that holds an operation together — who needs a check-in, what the last shift flagged, which document has the answer — lives in passing conversations, notebooks, and text threads. Boots was built to carry that load in one continuous loop.
Four systems, one platform.
Tenant isolation & administration
A multi-tenant foundation — organizations, brands, locations — enforced in the database with row-level security on every tenant-scoped table, an append-only audit trail, real-tenant provisioning, role-gated settings, and an operator console for running the product across locations.
Document retrieval & citations
Business documents are ingested, chunked, embedded, and indexed per location. Chat answers retrieve only the asking location’s documents and cite their sources — and the system is built to say so when the documents don’t support an answer.
Voice, bilingual & persisted workflows
A voice-first program (PWA shell, spoken Read-In, hands-free Q&A, conversational voice Read-Out), English/Spanish bilingual chat, phone sign-in for frontline workers, push nudges, and a KPI dashboard — with the daily loop persisted in the database, not a browser session.
Since the rebrand
As Boots, the platform kept moving toward the general small-business owner: assistant profiles, booking and appointment workflows, a Boots demo organization, and continued security-policy cleanup — seven further database migrations on top of the original build.
Three decisions, with the receipts.
Every reference below is a dated pull request in the platform’s delivery record — the decisions a senior engineer would ask about, answered with evidence.
Tenancy enforced by the database
Every tenant-scoped table carries row-level-security policies keyed to the active organization, so a query physically cannot return another tenant’s rows — the boundary never depends on a developer remembering a filter. The pattern was ported from operator-provisioning work already proven at enterprise scale.
PR #5 · May 11, 2026 · Migrations 001–009A grounded answer — or an honest one
Retrieval is scoped to the current location, responses carry source citations, and when retrieval or a provider fails, the system degrades gracefully instead of bluffing. When the embedding model changed the math, the retrieval threshold was recalibrated against it.
PR #25 · Jun 3, 2026 · PR #95 · Jun 10, 2026Persistence failures are P0
When a chat-persistence defect surfaced that could lose messages, it was treated as a P0 data-loss bug and fixed the same day it was found — on a Sunday, at the end of a ten-hour build day that also shipped the bilingual chat release. A system people trust with their daily record cannot quietly drop it.
PR #76 · Jun 7, 2026Dated, reproducible, and bounded.
| Measure | Build record · Jun 25, 2026 | Boots repository · Jul 2026 |
|---|---|---|
| Commits | 160 — 157 authored by Damara (98%) | — |
| Merged pull requests | 107 — 104 authored by Damara (97%) | — |
| Database migrations | 48 | 55 at Jul 23 — the original 48 plus 7 added since the rebrand |
| Application code | ~33,000 lines of TypeScript | — |
| Design corpus | 88 documents · ~195,000 words of specs, plans & handoffs | Specs and handoffs continue with every program |
| Automated tests | — | 498 of 498 passing (52 files) at the Jul 12, 2026 full-stack audit, with lint, typecheck & build gates green |
| Active build days | 27, including 16 weekend commits | — |
Every figure reproduces from the repository’s Git history, pull-request record, or migration folder, as of the date shown. Two honest bounds: live HRIS/scheduling integrations were deliberately not built — the specification gated them on an operator confirming their stack, and speculative integrations are how AI projects burn budget; the CSV fallback and integration registry shipped instead. And these counts measure delivery, not business impact — what Boots does for a business’s retention or operations is a claim only real usage data can support, and this page doesn’t make it.
Need an AI product built like this?
Start with a Grounded AI Blueprint — the fixed-scope first engagement on our pricing page: a documented assessment of your workflow or prototype, prioritized risks, a recommended architecture, and a phased implementation plan — before you commit to a larger build. Any build that follows is a separate scope.
Straight answers, same as the build.
Who designed and built Boots?
Damara Carlentine, founder and principal of Boots On The Ground AI, LLC. She wrote the requirements, specifications, and architecture; directed AI coding tools under her own plans and review; authored the database migrations; and delivered the features — 157 of 160 commits and 104 of 107 merged pull requests in the build record are hers.
What problem was Boots built to solve?
Operational knowledge in frontline businesses lives in people’s heads, notebooks, and text threads — and walks out the door with turnover. Boots gives a business a persistent, document-grounded AI teammate that briefs the day, answers from the business’s own documents with citations, and captures the day’s record for whoever comes next.
What do the delivery figures actually prove?
Delivery: that a complete, multi-tenant, production-grade AI platform was designed and built, with the receipts to verify it. They do not claim adoption, revenue, or retention outcomes — no such data is asserted anywhere on this page, deliberately.
Can a system like this be built for our workflow?
That’s what the Grounded AI Blueprint — the first engagement on the pricing page — is for. Boots reviews one agreed workflow or prototype and returns findings, a recommended architecture, and a phased implementation plan — usable with your own team, or as the basis for a separately scoped build with Boots.