MICROSIVE
← All work
SaaSAI/ML2024

Kestrel AI

Shipped a B2B SaaS MVP in 7 weeks. 3 pilot deals closed in month 2.

Aman had a working prototype — a document intelligence tool that let ops teams query their contracts, SOPs, and vendor agreements in plain English. He'd been demoing it for three months. Every demo got a 'yes, we'd pay for this.' None of it was production-ready.

$180K

ARR in 90 days

7 weeks

prototype to production

3 deals

closed in month 2


01 / Problem

A great demo. Nothing behind it.

The prototype was a Flask app running on Aman's laptop. RAG pipeline worked, but the document ingestion was manual, there was no multi-tenancy, no auth, no usage tracking, no billing, and the latency was 8 seconds on average. He needed a production system fast — two pilot customers were waiting and a third was getting impatient. He'd been quoted 6 months and ₹40L by two other agencies. He had 8 weeks and ₹15L.


02 / Solution

Ship the thing that closes deals. Nothing else.

We spent the first week deciding what not to build. The first version needed: working multi-tenant auth, document upload + processing pipeline, query interface, and usage metering for billing. That's it. No settings page, no team management, no API docs. Those came later. We rebuilt the RAG pipeline on a production stack, added a proper document processing queue, and shipped a clean query UI that didn't look like a prototype.

  • Multi-tenant auth with organisation-scoped data isolation
  • Document ingestion pipeline: PDF, DOCX, CSV processing queue
  • RAG pipeline rebuild: pgvector + OpenAI embeddings, latency 8s → 1.2s
  • Clean query interface with source citation and confidence indicators
  • Usage metering via PostHog events → Stripe billing integration
  • Admin dashboard for Aman to manage organisations and usage
  • Weekly Friday demos from week 1 — pilot customers watched it being built

03 / Result

Three paying customers before the formal launch.

The Friday demo rhythm turned out to be a sales tool. Aman started inviting pilot prospects to join the weekly calls. By week 5, two had signed LOIs. The third signed after the production launch in week 7. The speed of the build also changed how investors perceived the company — 'they shipped in 7 weeks' became part of the pitch deck.

  • $180K ARR within 90 days of production launch
  • Query latency: 8 seconds → 1.2 seconds average
  • 3 pilot deals closed in month 2
  • Document processing: manual → fully automated ingestion
  • Zero downtime in first 60 days of production

04 / Deliverables

What they walked away with.

  • Production SaaS application (Next.js + FastAPI)
  • RAG pipeline with pgvector + OpenAI
  • Multi-tenant auth system
  • Document processing queue
  • Stripe billing integration
  • Admin dashboard
  • GitHub repo (theirs), full documentation
  • Notion handover + architecture overview

05 / Stack

The tools we used.

Next.jsPython (FastAPI)PostgreSQLpgvectorOpenAI APIPineconeStripePostHogVercelRailway

The only agency that's ever shipped on the date they promised. We were so suspicious we ran the build ourselves. It worked.

AS

Aman S.

Founder, Kestrel AI

Next case study

↑ 3.4× daily active use

Midori Health

Built an offline-first care tracker used by 800 clinicians.

Want results like these?

Tell us what you're building.

Start a project →