Building both sides of a modern brokerage.

The Mortgage Panel, run by Lloyd Raphael, manages 406 client files, 792 deals and 36 lenders. The practice is a willing reference on request, and its public website is at themortgagepanel.com.au.

Every loan application generates a large volume of identity documents, payslips, bank statements, contracts and lender correspondence. Most of it arrives by email, out of order and under whatever filename the sender chose.

All of it lives in the practice's own Microsoft 365 tenant, and it has to stay there. Any assistant had to work inside that boundary, under an identity the practice controls, rather than copy client documents anywhere else.

Filing those documents, finding information when a lender or client asked for it and keeping clients updated was consuming most of a personal assistant's week. The practice did not need another chatbot. It needed the administrative layer of the business to run itself, with a person still in control of anything that mattered.

It also needed to win clients. The practice's public website had to be found for the right searches, answer real questions, and hand a ready client to the broker with the details already in the CRM, in a sector where every claim, disclosure and form is subject to compliance review.

Most vendors own part of the stack. A web agency understands the website. An AI automation agency understands a model and a trigger. A managed service provider understands infrastructure and support. This engagement needed all of it to work as one environment.

Sigma Labs designed and built the client-acquisition layer, the public platform that finds, informs and qualifies a customer and hands them to the broker, and the business-operations layer, the private AI assistant and specialist agents that do the practice's document, email, contract, research and reporting work under human approval. Both run against the same Microsoft 365 tenant, the same CRM and the same security boundaries.

Financial-services platforms cannot be treated like ordinary marketing sites. They need appropriate attention to advertising claims, customer disclosures, privacy and consent, web forms, AI and question-and-answer content, regulated terminology, lead capture, content accuracy, ongoing review, security, accessibility and user experience.

The Mortgage Panel website was reviewed by AFG Compliance and approved for use. The approval specifically required the brokerage to continue using Sigma Labs as its provider and to have the platform reviewed regularly.

That is a review of The Mortgage Panel's website, not a certification of Sigma Labs, and it does not make compliance a guarantee. What it shows is that the platform was engineered and worded with the regulated context in mind, and that a compliance function saw ongoing Sigma Labs review as part of keeping it that way. Regulated figures and scheme names on the site are held in one fact model with a source and a verification date, so a change is made once and reviewed rather than hunted across pages.

Sigma Labs designed and engineered Mia, the practice's private AI operations assistant, integrated into the brokerage's working environment: its mailboxes, document libraries, client records and lender information.

Mia communicates through a branded web console, Telegram messaging and a morning brief, with a WhatsApp channel in progress. Behind those interfaces it coordinates nine specialist agents, each scoped to a single job: inbox, filing, correspondence, lending research, proposals, income analysis, serviceability, diary and contract review. On this page they are grouped into the six jobs the practice actually sees.

The AI is the visible part. The larger part of the work was architecture: the identity Mia acts under, the permission boundaries around the documents, the approval gates, the audit trail and the console that shows what it is doing and what it costs.

Mia's name, voice and interfaces are specific to The Mortgage Panel. The architecture, agent orchestration, security boundaries and approval controls underneath are Sigma Labs engineering, shaped around each organisation we work with.

Mia is a human-in-the-loop system by architecture, not by policy. Sending an email, filing a document to a client record, issuing contract findings and any other action with a consequence for a client, lender or the practice waits for a person to review and approve it.

Approval gates are enforced in the workflow code and visible on the console. Lower-risk work such as classification, drafting, summarising and research runs ahead, so the person's time is spent on decisions rather than on preparation.

A new client in one conversation: folder, File Tracker and intake draft created, then Mia stops to confirm the lead details and asks for the one field she does not have before creating it. Names redacted.

Built to be trusted

• Drafts, never sends. Every client-facing email, lead and new folder needs the broker's yes.
• Reads before it names. Filenames are never trusted; every document is classified from its content and mismatches are flagged.
• The filing standard is enforced, not requested. A file name that breaks the standard is rejected by the platform before it is saved, and a guard sweeps the whole library six times a day, repairs what it can and only alerts a person for what it cannot.
• Uncertain files are tagged for review rather than guessed.
• Nothing is overwritten or deleted. Client documents and records are stored only in the practice's own Microsoft 365 tenant; model processing is documented and controlled.
• Partial reads are disclosed. If a document was only partly read, the broker is told exactly which pages were missed.

The filing standard polices itself: a rejected name with the reason, and the library guard's runs. Client names redacted.