Collins Kiptoo
Collins Kiptoo, Software Engineer. Full‑stack, backend‑first, applied AI.

I build complete products, from the first screen to the last API call.

Five years shipping production software for startups and growing companies: a lending platform used by 6,000+ people, a freelance marketplace with escrow payments, an e-commerce tool that keeps five sales channels in sync, document and image AI, SaaS dashboards and booking apps. I own the hard parts: data models, integrations, permissions and edge cases. Lately, AI agents that have to show their work.

Available now for contract or full-time roles. Remote from Nairobi (UTC+3): full overlap with Europe and the UK, mornings with the US East Coast, afternoons with Australia.

Collins Kiptoo
SHIPPED SINCE 2020
production software
Lending platform, 6,000+ users2024·✓
Asset-finance dashboard2023·✓
Document-reading AI2023·✓
Freelance marketplace, escrow2022·✓
5-channel e-commerce sync2022·✓
Image AI, pest counting2022·✓
Booking apps, SaaS, tools2021·✓
Products shipped15+
5 yrs shipping to production
15+
products shipped to production since 2020
30+
third-party integrations in a single platform
6,000+
users on the largest product I worked on
5.0
from every client who rated my freelance work
What I'm known for

Three things people come to me for.

Building the whole product, not one screen of it.

Data model, backend workflows, the interface, the admin tools behind it and the emails it sends. I know how a change in one shows up in the rest.

What I've built

Connecting a product to everything around it.

Payments, marketplaces, CRMs, accounting, e-signature, email, maps, credit checks and AI models. Real API calls with authenticated webhooks, not just plugins.

30+ in one platform

Getting the hard parts right.

Money that has to add up, data that has to survive an audit, and a live platform rebuilt on Google Cloud without losing a customer. The engineer new hires get pointed at.

How I work on Bubble
Selected work

The receipts.

Different products, different industries, same standard. Client names and amounts stay private where they need to; the counts are exact.

End to end a lending platform, 6,000+ users

Every step of a product, from the installer's quote to the final payout.

At Cloover I worked on every part of a lending platform for solar, batteries and heat pumps across four countries: the offers installers build, checkout, the credit decision, contracts, invoices, payments, and 30-plus integrations underneath. When it was rebuilt on Google Cloud, I was the team's source of truth for how the old system behaved: I found and fixed the parity defects and specified the behaviour, while the Google Cloud engineers built the new services.

Read the case study
OFFER OF-2291
solar, battery, heat pump
Solar PV, 9.8 kWp0% VAT·✓
Battery, 10 kWh0% VAT·✓
Heat pump19% VAT·✓
Credit checkSchufa·✓
Affordability rulepassed·✓
Contractsigned·✓
StatusFinanced
Illustrative offer
Two-sided a freelance marketplace

A marketplace where both sides trust the money.

Clients browse packaged projects with fixed scope and price, sign a contract, and fund each milestone into escrow. When the work is approved, the platform takes its fee and pays the contractor. I built it end to end: Stripe Connect split payments, escrow-protected milestones and DocuSign contracts.

See it with the others
PROJECT #RM-118
two-sided marketplace
Scope and price agreedlisting·✓
Contract signedDocuSign·✓
Milestone 1 fundedescrow·✓
Work approvedclient·✓
Platform fee splitStripe Connect·✓
Contractor paidpayout·✓
MilestoneReleased
Illustrative example
5 channels one product, listed everywhere

Sell once, update everywhere.

An e-commerce manager that imports a product once and lists it on Shopify, Amazon, eBay, Facebook Marketplace and Mercari. When it sells on one channel, stock updates on the other four. Five different APIs, five sets of rules, one source of truth.

See it with the others
SKU 22-0417
one listing, five channels
Shopifylisted·✓
Amazonlisted·✓
eBaylisted·✓
Facebook Marketplacelisted·✓
Mercarilisted·✓
Sold on eBaystock 12 → 11·✓
Every channelIn sync
Illustrative example
No typing document-reading AI

Invoices that read themselves.

Koncile turns scanned invoices into structured data and matches each line to its purchase order, so finance teams stop typing. I built the product people use: the front end and the review flows that send anything doubtful to a human, on top of an AWS Textract extraction pipeline.

More on my AI work
SCANNED INVOICE
read, structured, checked
Supplier and dateextracted·✓
14 line itemsextracted·✓
Matched to purchase order13 of 14·✓
Price mismatch on line 9flagged·✓
Sent for human reviewqueue·✓
Manual entryNone
Illustrative example
A workflow that's 98% right is a workflow somebody fixes by hand every morning.
The sentence I design against
AI, honestly

I use agents every day. I trust them with nothing I can't check.

  • Public now: M-Pesa Statement Agent.FastAPI and a typed LangGraph workflow that categorises mobile-money statements by rule, flags duplicate charges, spending spikes and failed reversals, and answers grounded questions. Tests and CI, runs with no API key. A demo on synthetic data. Read the code
  • In daily use since 2025.Claude Code and Buildprint inside a large production app. They surfaced a security issue in a partner integration and traced a refund bug that was quietly marking refunded invoices as paid.
  • Building now: an agent that matches payments to invoices.Rules first, the model only for the messy remainder, a human for anything it isn't sure about. It goes public with its evaluation numbers, or it doesn't go public.
  • The rule I work by.The agent proposes. The data decides. A match nobody can explain counts as a miss.

More on how I use AI, and what I won't claim   My GitHub

AGENT-ASSISTED, IN PRODUCTION
Claude Code and Buildprint, 2025 to 2026
Security issue in a partner webhookfixed·✓
Refunds being marked as paidfixed·✓
Fee errors across every live project7 batches·✓
Field missing on rebuilt platform289 refs·✓
Duplicate invoices, root causefound·✓
Every outputChecked by hand
How I got here

From actuarial maths to shipping software.

  • 2020

    Insurance data, as an actuarial intern

    Validating data for risk pricing at Jubilee Insurance. Where I learned that a number is only useful if you can say where it came from.

  • 2021

    First clients, on Bubble

    Freelance builds for early-stage founders: booking platforms, internal tools, a QA audit written up as a report.

  • 2022

    Six apps at once, for a US agency

    A marketplace with split payments and escrow, an inventory sync across five sales channels, an app that counts pests from photos.

  • 2023

    Product studio work, Berlin

    Fence, a dashboard for lenders that later raised a $20M Series A; Koncile, a document-reading AI tool; and Aspire, a construction planning app.

  • 2024

    A full platform at Cloover

    Every part of a lending platform used by 6,000+ people: offers, checkout, credit checks, contracts, invoicing, payments and 30+ integrations, then its rebuild on Google Cloud.

  • 2026

    Agents that show their work

    Taking everything above into AI systems that are measured before they're trusted.

What I work with

Tools I use in production, and what I'm learning now.

In production

Bubble.io (5 years), JavaScript and Node.js, SQL, REST APIs and webhooks, Stripe and Stripe Connect, SEPA, Shopify, Amazon and eBay APIs, HubSpot, QuickBooks, DocuSign, DocuGenerate, Postmark, Zapier, Google Cloud, Cloudflare, AWS Textract, Xano, Supabase, FlutterFlow, Figma

AI tooling

Claude Code and Buildprint daily since 2025, Claude API, OpenAI API, MCP, retrieval (RAG), evaluation harnesses for LLM output

Learning now

Python, TypeScript, React, Docker. Practising every day, with the work going up on GitHub as it ships.

Got something to build?

collo060@gmail.com
Collins KiptooNairobi, on European hours.
Contract, full-time, or a short paid pilot.

Email Collins

collo060@gmail.com

Open in Gmail Open in Outlook Use my mail app