Most AI projects stall at the demo. Ours ship.
The hard part of AI in a real business is rarely the model. It's the integration, the data underneath it, and the question of who owns the thing once it's live. That's the part we build for.
About Us
We build AI automation that keeps running after we leave. Built on the Microsoft stack, documented, and owned by your team.
- Production AI automation on Azure: Functions, Logic Apps, Cosmos DB
- LLM agents wired into the tools a team already uses, with human approval gates
- Documented handover: runbooks, re-run procedures, and a 30-day onboarding plan
We are a small, senior team. That means you work directly with the people building your system, not a layer of account managers above them. We take on engagements where the goal is a working system someone else can operate, so every build ships with the documentation, failure procedures and onboarding plan needed to hand it to your own hire. We put that in writing at the start, and we hold ourselves to it at the end.
See our services-
01 Why do AI pilots fail to reach production?
Usually for three reasons that have nothing to do with the AI: no one owns it after launch, there's no runbook when it breaks, and there's no defined path for what happens when a step fails. A demo only has to work once. A production system has to work on a Tuesday morning when the person who built it is unreachable.
-
02 What makes our builds different?
Human-in-the-loop by design. Anything that goes out under your name, whether an email, a message, or a commitment to a customer, is drafted by the system and sent by a person. The automation does the reading, scoring and drafting; a human keeps the final click. That gate is deliberate, and it's what keeps your brand voice and your relationships under your control.
-
03 What do we leave behind?
An architecture document, a runbook covering what runs on its own and what needs a person, manual re-run procedures for the failures most likely to happen, a monthly operating checklist, and a named list of what we'd do next. Enough that a new hire can pick it up in their first month.
What we're actually good at
A narrow stack, worked in depth, rather than a long list of logos.
- Azure architecture: Functions, Logic Apps, Cosmos DB, Service Bus
- LLM agents and Slack or Teams assistants your team talks to directly
- Data consolidation and integration layers across disconnected tools
- Handover documentation, runbooks and operating checklists
Services
Four ways we work, from a two-week look at your operation to a system running in production.
AI Strategy & Roadmap
Where AI actually pays off in your operation, and where it doesn't. A prioritized roadmap with effort, risk and sequencing.
AI Automation Build
Agents and automations wired into the tools you already use, with a human approval gate on anything that goes out in your name.
Data & Cloud Foundation
Consolidate scattered tools onto one Azure environment with a single source of truth, so AI has something reliable to read.
Handover & Support
Runbooks, re-run procedures and an onboarding plan, so your team can run it without us. Plus a retainer if you'd rather not.
Not sure where AI fits yet?
That's the usual starting point. A short assessment gets you a ranked list of what's worth building, and what isn't.
Case Studies
The same architecture and the same human approval gate, across three very different operations.
Recruiting & staffing · Delivered
From intake form to booked interview, with a person on every send
5 → 1systems consolidated into one source of truth
7pipeline steps automated end to end
6failure paths with written re-run procedures
40-60%targeted cut in tool and maintenance spend
The problem A US recruiting firm was running on a scattering of disconnected tools. Job data lived in one system, candidates in another, email in a third, and nothing reconciled between them. Every step of the pipeline depended on somebody remembering to do the next one, so work fell through exactly where it cost the most: a candidate who replied with interest and never heard back.
What we built Email and DNS moved onto Microsoft 365 first, so there was one identity and one mailbox to build against. Then a single database holding every job, candidate and reply, as the source of truth. On top of it, three automation services covering job intake, candidate sourcing and inbound email; two pipelines connecting the public intake forms to the system; and an integration layer so the existing recruiting tools wrote into the same place instead of into their own silos. Day-to-day control lives in Slack, through three assistants for logging a role, looking up a candidate, and approving outbound email.
The result A seven-step pipeline now runs from intake form to booked interview. Inbound replies are read, classified and scored without anyone touching them. Every outbound email is still drafted by the system and sent by a person, because the firm's reputation rides on what goes out under its name. We handed over a runbook covering what runs on its own and what needs a human, manual re-run procedures for the six most likely failure points, and a 30-day onboarding plan written for the hire who would inherit the system.
Accounting & professional services · Delivered
Client documents that file themselves, replies that still come from you
4 → 1intake channels consolidated into one queue
3attributes read off every document: client, type, period
100%of client-facing email released by a person
The problem Client documents arrived by email, portal, text and hand. Someone opened each one, worked out which client and which period it belonged to, renamed it and filed it. In busy season that job quietly became a full-time role, and the answer to "did we get their W-2 yet?" lived in one person's memory.
What we built Every intake channel now lands in one queue. An agent reads each document, identifies the client, the document type and the tax period, files it against the right engagement, and updates a live checklist of what is still outstanding. Chase-up emails are drafted automatically and queued for a partner to send.
The result Nothing addressed to a client, and nothing touching a filing position, leaves without a named person approving it. The automation took the sorting off the firm's desk and left the professional judgement, and the liability, exactly where it belongs.
Property management · Delivered
Maintenance requests triaged and dispatched before anyone opens the inbox
24/7triage on inbound tenant requests
2drafts per request: vendor dispatch and tenant reply
0vendors dispatched without a manager approving
The problem Tenant requests arrived at all hours across email, phone and a portal, in wildly varying detail. A manager read each one, decided whether it was an emergency, worked out which vendor covered it, and started a thread. Urgent items sat behind routine ones purely because of the order they arrived in.
What we built Requests are read on arrival, classified by trade and urgency, matched to the property, the unit and its repair history, and ranked. A leak outranks a squeaky door without a person deciding that at 11pm. The system drafts both the vendor dispatch and the tenant acknowledgement, ready to send.
The result A dispatch commits money and a tenant reply sets an expectation, so the system prepares both and a manager releases them. Nobody wakes up to a vendor booked by software, and nothing urgent waits until morning to be spotted.
Client names, systems and identifying detail are withheld at our clients' preference. Happy to talk through any of these in more depth on a call.
Our Team
The people who build your system are the people you talk to.
Cherie Topham
Co-FounderAzure architecture and LLM integration. Designs and builds the automation, and writes the documentation that goes with it.
Anthony Blanksteen
Co-FounderData platforms and integration design. Focuses on making scattered systems behave like one, and on keeping them reliable.
How We Engage
Three shapes of work. Most clients start with the first and decide from there.
AI Readiness Assessment
Fixed scope · 2–3 weeks- Review of your current tools and data
- Ranked list of AI opportunities
- Effort, risk and sequencing for each
- A roadmap you own, whether or not we build it
Implementation Sprint
Scoped build · fixed fee- Everything in the assessment
- Built and deployed on your Azure tenant
- Human approval gates where they matter
- Runbook and handover documentation
Ongoing Support
Monthly retainer- Monthly health review
- Config drift checks
- Roadmap items delivered monthly
- A named person to call
Frequently Asked Questions
The questions we get asked most often before an engagement starts.
-
Do we need to be on Azure already?
No. Moving you there is often step one. A common starting point is a business running on a scattering of disconnected tools. That consolidation is usually what makes everything after it possible, because AI is only as good as the data it can reliably read.
-
Will AI be sending emails as us?
Not without a person clicking approve. We build the drafting, reading and scoring as automation, and we put a human gate in front of anything that leaves your organization. You can widen that gate later once you trust the output, but it starts closed.
-
What happens when you're done?
You get a runbook covering what runs on its own and what needs a person, manual re-run procedures for the most likely failures, and a 30-day onboarding plan written for whoever inherits the system. If you'd rather we stayed on, there's a monthly retainer, but the handover is built so you don't need one.
-
How long does a first project take?
The assessment is two to three weeks. A first automation running in production is typically six to ten weeks after that, depending on how much consolidation has to happen underneath it. We'd rather tell you the honest number up front than discover it halfway through.
-
What if we already have an AI tool that isn't working?
Common, and usually fixable. In our experience it's rarely the model. It's an integration or data problem. The tool can't reach the systems it needs, or what it reads is incomplete. The assessment will tell you which of those you have before you spend anything replacing it.
Contact
Tell us what your operation looks like today and where it's slowing down. We'll tell you whether AI is the answer.
Location:
Las Olas Blvd, Fort Lauderdale, FL 33301
Email:
info@lasolasdev.com
Call:
+1 305 409 6993