KAVYCorp
 Get in touch
✦ Co-founder · KAVY CORP · Cali, Colombia

Diego
Burbano
ships it.

Full-stack developer, programming since 2020. I design, build and run production software on my own, from the database to the WhatsApp message: warehouse systems, payment checks, booking systems and AI tools. Open to remote roles.

See the work LinkedIn
30+people use the warehouse app I built at SILOG
299unit tests on the money rules of a booking system
C1English · Spanish native · UTC−5
Selected work

Five systems, built end to end.

Each one is running or deployed. Where something hasn't gone live yet, the card says so. Code is private; I'm happy to walk through it on a call.

01 · Own product

KAVYCHECK

Tells a shop owner, in seconds, whether a bank-transfer screenshot sent on WhatsApp is money that actually arrived.

Live · used in my own business
Screenshot on WhatsApp→OCR→Match bank email alert→Verdict
kavycheck.app ↗

Engineering decisions

  • Email forwarding instead of Gmail OAuth. Reading Gmail is a restricted scope with a yearly security audit and 7-day tokens; a forwarding rule is simpler for the customer and cheaper to run.
  • Official WhatsApp Cloud API only. Unofficial libraries risk a number ban, which a paid product can't afford.
  • The screenshot is never trusted. The source of truth is the bank's email alert; the parser handles both Bancolombia templates (numeric key and Bre-B) and checks DKIM.
  • Billing with three safety nets: Wompi webhook, reconciliation and a scheduled job, tested with a real payment.
Next.js 16React 19SupabaseDenoWhatsApp Cloud APIWompiOpenAI (OCR)

Self-serve sign-up is open. No external paying customers yet.

02 · Developer at SILOG

Warehouse management app

The system a free-trade-zone logistics company uses to receive, store, count and dispatch goods, on Honeywell handhelds and the web.

In use · 30+ users
Receive→Label→Put away→Count→Dispatch
Product page ↗

What I built

  • Pallet labels printed natively in ZPL to the warehouse Zebra printer, plus Bluetooth printing straight from the browser (Web Serial).
  • Delivery notes read with vision AI and mapped to the catalog through an equivalence table.
  • Receiving is atomic and idempotent, so a dropped connection on the warehouse Wi-Fi can't double-count stock.
  • Security in the database: every RPC function checks who is calling it, anonymous access revoked, per-module permissions per user.
  • Units of measure with decimals (kg, lb, units, m², m³), truck photo records and an editable stock ledger with audit trail.
React NativeExpo / EASSupabasePostgreSQL · RLSZPLWeb SerialClaude (vision)

Built for my employer; no screenshots or company data shown here.

03 · Client project

Padel club booking & cash

Bookings, split payments, refunds and daily cash closing for a six-court padel club. Built in two weeks.

Deployed · on-site go-live pending
Book a court→Players pay their share→Refunds→Close the till

Engineering decisions

  • Payment status is derived, never stored. A booking has a total; payments are rows from each player, so paid, partial or pending can't drift out of sync.
  • No overpayment, enforced by a database trigger, so the rule holds even if someone skips the app.
  • 299 unit tests on the money rules. Testing caught two real bugs before handover: cancelled bookings counted as revenue, and partial payments reported as $0.
  • Live calendar with Supabase Realtime; staff accounts managed through server actions, with the service key kept out of the browser bundle.
Next.js 16SupabaseRealtimePostgreSQL triggersTailwindVercel

Client system; screenshots available on request with demo data.

04 · Own tool

Postula

A job-search assistant I built for my family: it interviews you about your results, then tailors a CV, cover letter and salary answer to each job post.

Live · private app
Upload CV→Achievement interview→Paste a job→Tailored CV

Engineering decisions

  • It must never invent. Every line of a tailored CV has to trace back to the person's profile; limits like "this stayed a pilot" are stored so no CV can contradict them.
  • Audited adversarially: independent reviewers checked 7 real tailored CVs against the profile; high-severity issues went from 47 to 0 after one prompt revision.
  • Prompt caching on the profile cut the cost of each interview answer by about 70%.
  • Answers are tied to the question on screen with a conditional update, so a stale tab can't attach a number to the wrong job.
Next.js 16React 19PostgreSQL (Neon)Claude APIStructured outputsVercel
05 · Client automation

Nightly ERP → store sync

Keeps a Shopify store aligned with the company's accounting system (Siigo) every night, without anyone touching it.

Running nightly since Sep 2026
Siigo API→Compare→Fix prices & ghosts→Shopify API

What it does

  • Aligns prices with the ERP, zeroes products that no longer exist there, and guards that every live price is in pesos.
  • Safety brake: if a run would change more than 60 items, it changes nothing and flags it.
  • The first applied run cleaned euro compare-at prices from 17 live variants and left the catalog with 0 price mismatches.
PythonGitHub ActionsShopify Admin APISiigo API
How I work

Small team energy, production habits.

Most of what I've built, I've built alone, so I own every layer and every consequence.

Rules live in the database

Money limits, permissions and caller checks go in triggers and policies, not only in the UI.

Verify on the real thing

A green build isn't done. I test on the device, the printer or the production URL before calling it shipped.

Honest status

Pilot is pilot, pending is pending. I'd rather say it than have it found out.

AI-assisted, human-owned

I build with AI coding assistants, and I set the architecture, review every change and own what ships. I also measure what AI calls cost and cut it.

Background

Engineer first, then everything else.

Education · since 2020

B.Sc. Electronic Engineering & B.Sc. Multimedia Engineering

Universidad de San Buenaventura, Cali. Java, Python, C++, C#, React and Unity along the way. Academic Excellence award in 2022 (2nd in the Multimedia program, GPA 4.67) and an IoT smart-bin project at the IEEE CAS Student Design Competition 2021–2022.

Research · 2024–2025

Young Researcher, MinCiencias

Classified emotional states from EEG with a Random Forest: 83–84% accuracy, macro F1 0.82, plus a real-time dashboard in Python. Pipeline on GitHub ↗

Certificates

Prompt Engineering, Scrum, English C1

Prompt Engineering course (Platzi, 2026) · Scrum Foundation Professional Certificate (CertiProf, 2023) · EF SET English C1, 65/100 (2025). Before tech roles: led an Odoo ERP rollout and ran paid ad campaigns.

Let's talk.

Remote roles, contract work or a code walkthrough of anything above.

dfelipeburbano@gmail.com LinkedIn WhatsApp