MakeItMakeSense

The consistency layer for founder document to catch every conflicting number across your pitch deck, one-pager, and board memo before an investor does

Comment

Inspiration

Every founder has been there — the night before an investor meeting, five tabs open, realizing their ARR is $2.1M in the pitch deck and $2.4M in the one-pager. They fix it manually, at midnight, hoping they caught everything.

The problem isn't carelessness. It's that no tool treats founder documents as a connected system. Every file is an island. We built MakeItMakeSense to connect them.

What it does

MakeItMakeSense is a business document workspace with three things:

Cross-document intelligence — on every save, MakeItMakeSense extracts financial metrics from all your documents and compares them. Conflicts get flagged with a badge. One click syncs the correct value everywhere with a full audit trail.

Math verification — checks if numbers are internally correct, not just consistent. If cash is $3.2M and burn is $192K but you wrote runway as 24 months, MakeItMakeSense catches it.

AI document agent — drafts documents grounded in your actual company numbers. It uses what's already in your workspace and never invents figures.

Discord trigger via OpenClaw — trigger a full document audit from Discord. MakeItMakeSense scans every doc, resolves conflicts, and posts a structured report back to the channel.

How we built it

  • Frontend: Next.js with TipTap as the rich text editor
  • Claim extraction: Claude API extracts canonical financial metrics as structured JSON on every meaningful save
  • Conflict detection: deterministic comparison across every document pair with per-metric tolerances
  • Math verification: second Claude call checks internal consistency across all extracted metrics
  • AI agent: LangGraph StateGraph cycling through retrieve, plan, draft, check, emit
  • OpenClaw: Discord bot that routes commands to the MakeItMakeSense API and posts results back to the channel
  • Storage: Supabase with pgvector, row-level security on every table

Challenges we ran into

Founders write numbers in too many formats — $2.1M, 2.1 million, $2,100,000. Normalizing all of these reliably into base units required more prompt engineering than expected.

Defining when to run analysis was harder than the analysis itself. Every keystroke was too noisy. We built a debounce and diff-size gate so conflict detection only fires on financially meaningful changes.

The agent kept inventing plausible-sounding numbers. We solved it by running structured JSON extraction first and passing only verified claims to the drafting step.

Accomplishments that we're proud of

Detecting six distinct conflicts across three documents and resolving all of them in under ten seconds. The math verification catching an arithmetic error that no cross-document check would ever find. Cutting five half-built features at 2am and shipping three well-built ones instead.

What we learned

The hardest part of building AI tools for professionals is deciding what the AI should never do. The moment the agent invented a number that sounded right, the whole product became untrustworthy. We learned to treat a blank response as strictly better than a confident wrong answer.

What's next for MakeItMakeSense

  • Data room mode: one-click export of all workspace documents, conflict-free and math-verified, ready to share with investors
  • Live connections to Mercury, Stripe, and QuickBooks so metrics come from the actual source of truth rather than being typed manually
  • Team collaboration with invites, roles, and real-time co-editing for Series A and B teams
  • Industry-specific claim schemas for SaaS, marketplace, and e-commerce business models

Built With

Share this project:

Updates