Projects /crm-forge
CRM Forge
An AI-first CRM for small sales teams. The assistant finds prospects and shows you the page each one came from, fills in a company record from the firm's own website, finds the people who work there, and drafts the email, and nothing reaches your workspace or your prospect until you press the button.
Why this exists
I wanted a CRM where the research is done for me and I can still check it. Every CRM I tried was a very nice spreadsheet, and every AI prospecting tool I tried was a black box that produced a list of names with no way to see where they came from. So I built one where the assistant does the legwork and has to show its work.
Ask it for prospects and it comes back with companies that fit, each with the page it read them on. A source only counts if the run's own web search actually returned it, so it cannot cite a page it never saw. Point it at a company you only have a name for and it reads the firm's own site, proposes the website, industry, size, phone, address and the people who work there, and, if it finds the firm was bought by someone, says so and lets you file the record under the acquirer with the relationship named. Nothing is written until you accept it, and the model's own confidence never counts as permission.
How a day goes
Companies, contacts, leads, deals, follow-ups and a map of where your customers are, with the usual saved views and imports. When it is time to reach out, the assistant drafts the email from the record and you send it from your own mailbox, so replies land where your mail already lives. Unsubscribes, suppression and the compliance footer are enforced at the one gate every send passes through, so a person who said stop is not emailed again by a queued step that was already waiting. A dashboard shows what moved this week, and an assistant can read those figures and tell you what they mean, from your data only.
You pay for the AI as you use it, from a prepaid credit, with the price shown before every run and the actual cost after.
The stack
.NET 10 and Blazor Server on Azure, EF Core against Azure SQL, MudBlazor for the UI, Claude behind the research and drafting, Microsoft Graph and the Gmail API for sending from the user's own mailbox, ECharts for the dashboard, MapLibre for the map, and Stripe for the plans. It is multi-tenant from the first line: every workspace's data is isolated at the query layer and the tests that prove it block a release.
Built solo with heavy AI assistance, and I am its first customer, which is a demanding arrangement. The fastest way to find out a feature is wrong is to need it on a Tuesday morning.