A commercial real estate investment group came to me for a new website. Today, its partners can answer investor and client inquiries from 27,000 pages of deal history in under 45 minutes, without access to the sensitive financial information inside, all from their Claude accounts.
He’s become someone we trust for both advice and execution.
Rasheed is great to work with. He quickly understands how different parts of our business operate and identifies opportunities for improvement that we hadn't considered. His work has significantly improved our response times and given us better tools to evaluate potential investments. He's become someone we trust for both advice and execution, and I wouldn't hesitate to recommend him to any business looking to improve its operations.
CAMCO buys and manages commercial property. Every deal it has looked at arrives as an offering memorandum: a long PDF of rent rolls, financials, tenant names and terms. Over the years the firm had collected 475 of them, about 27,000 pages, in SharePoint, and because they hold highly sensitive information, only the two managing partners could open them.
The firm wanted its partners to answer investor and client questions themselves, without anyone seeing what only the managing partners should, and without anything sensitive leaving the firm's own systems.
Three rules every part of the system had to keep.
Originals open to the two managing partners only.
Run on the Microsoft 365 and AI subscriptions the firm already had.
Nothing reaches the partners until a managing partner approves it.
A new website, a map of where AI fits in the firm's work, and a first prototype of the site-selection tool.
The document system and the site-selection tool built, tested, refined and validated before anyone relied on them.
Teaching the firm to use what was built, and making sure everyone asks it the right way.
Since then, CAMCO's ongoing AI advisor: maintaining both tools, and adding new ones as the firm finds new questions.
An event-driven pipeline inside the firm’s own Microsoft 365 and Azure. Every memorandum is read, redacted and summarized automatically, and a managing partner approves it before any partner’s AI can see it.
No custom server / no vector database / no always-on AI infrastructure
The original memoranda, in their own site with role-based permissions.
An event-driven workflow starts on every new or changed memorandum; nobody runs it by hand.
Reads every page: running text, tables and the financial figures in them.
Azure AI Language and Azure OpenAI find and remove sensitive details, then write a standard deal summary: NOI, IRR, cap rates, financing terms and risks.
A managing partner reviews every sanitized document and summary. Only approved files move on.
Sanitized copies and summaries, in a separate site the other partners can open.
Partners sign in with Entra ID and MFA; the AI inherits their SharePoint permissions, so it can only reach the approved site.
Partners sign in with their existing Microsoft 365 accounts through Entra ID with MFA.
The AI reaches only the documents that partner may already open.
A managing partner approves every sanitized file before it is published.
Microsoft Purview logs access, sharing and security events, alongside approval records.
One-time, to process all 475 memoranda
For about 12 new memoranda a year
Custom servers, vector databases or always-on AI infrastructure
Everything runs on the Microsoft 365 and AI team subscriptions the firm already had.
Data centers and other heavy-power tenants need sites near transmission capacity. Prototyped with the team on the first visit and built out between visits, the site-selection tool runs on the website's backend: it overlays public grid data on a prospective acquisition and asks an AI model for a weighted score of its fit.
Used by the team in the field.
New tools as the firm finds new questions.
Claude Team/Microsoft 365/SharePoint/Microsoft Entra ID/Power Automate/Azure AI Document Intelligence/Azure AI Language/Azure OpenAI/Microsoft Purview/Next.js/Auth0/Mapbox GL JS/Supabase PostGIS