The AI reads the prices, code does the sums.
A language model can get a number wrong. So in the systems I build, the model only reads the amounts, and code adds them up.
INKDRAFT · OWN PRODUCT
From sales call to paid contract.
InkDraft is a SaaS product I founded and operate. It takes a deal from the sales call through proposal and contract to signature and payment. It is live on paid self-serve plans.
- Role
- Founder and engineer, from architecture to operations
- Status
- Live, paid self-serve plans
- Stack
- TypeScript · Next.js · Prisma · PostgreSQL with pgvector · Trigger.dev · Stripe
- Product
- inkdraft.io
The problem
Service businesses agree a deal in conversation and then have to turn that call into a proposal, a contract and a payment request. Each time the deal moves from one tool to the next, a price or a term can end up different from what was said.
What it does
- Generates proposals, service agreements, mutual NDAs and partnership agreements from a call transcript
- Writes proposals from the company's own documents and case studies and revises them on request
- Shares documents through review links where clients can comment and request changes
- Collects electronic signatures in a signing step built into the product
- Sends the client a payment page right after signing, for one payment or instalments
Why it can be trusted with contracts
- The model reads the amounts from the call. Code adds them up and checks that fees and payment schedule equal the total
- Every price line needs a quote from the call. Without one, the extraction runs again
- A consistency check looks for contradictions between sections before a proposal goes out
- AI requests go to providers configured for zero data retention, and no provider trains on customer content
- Signing locks the document version, so what gets signed is what was reviewed
Business model
Subscription plans in Swiss francs, plus a fee on each contract payment processed through Stripe Connect.
What carries over to client work
The same rules apply to any system where a business commits to what the AI produces: compute figures in code, require evidence for every extracted fact, check for contradictions, and keep customer data out of provider training.
Questions
Who is InkDraft for?
Service businesses that agree a deal in a call and then have to produce the proposal, the contract and the payment request.
How does InkDraft keep the prices right?
The model only reads the amounts from the call. Code adds them up and checks that the fees and the payment schedule equal the total, and every price line needs a quote from the call.
What happens to customer data?
AI requests go to providers configured for zero data retention, and no provider trains on customer content.
Can something like this be built for our company?
Yes. The same rules carry over to any system where a business commits to what the AI produces: figures computed in code, evidence for every extracted fact and a check for contradictions.
Read next
What happens to your customer data when you use AI.
Every AI request sends your data to a provider. In the systems I build, each request tells the provider not to store the data and not to train on it.
Need something like this built?