All projects
Felt cover for AlışBuddy

AlışBuddy

  • Finalist

An AI shopping assistant designed for older adults. Runs on 15 LangGraph agents and meets WCAG AAA and KVKK rules. BTK Hackathon'26 finalist.

private code

Award

Finalist at BTK Hackathon'26, with AlışBuddy.

My role: Solo: all of the logic and flows.

Problem

Many older adults in Turkey stay away from online shopping, and the project names fear of fraud as the main reason. Fake SMS messages and look-alike shop addresses are hard to spot on a small screen.

Approach

An assistant called Buddy listens to voice and reads photos and text, built on 15 LangGraph agents with Gemini. Every link and message goes through Google Safe Browsing, a look-alike domain test for 40+ Turkish brands and a search over 30 known scam patterns. The interface uses WCAG AAA sizes: 18 px text, 56 px touch targets and 7:1 contrast.

Result

AlışBuddy reached the final of BTK Hackathon'26. The repository records 42 of 42 security tests and 162 of 164 Playwright end-to-end tests passing. Memory and consent flows follow KVKK, including a way to delete what Buddy remembers.

Highlights

  • BTK Hackathon'26 finalist
  • 15 LangGraph agents
  • 42/42 security tests
  • 162/164 end-to-end tests
  • WCAG AAA sizing

Stack

  • Next.js
  • React
  • TypeScript
  • Tailwind CSS
  • LangGraph
  • Gemini
  • Postgres (Neon)
  • pgvector
  • Drizzle
  • Redis (Upstash)
  • NextAuth
  • Capacitor
  • Remotion
  • Playwright
  • Vitest
AlışBuddy voice question screen
The voice question screen with Buddy's six-step intro tour.
Fraud risk result for a fake SMS
From the demo video: a fake bank SMS gets an 84% fraud risk, with the domain, Safe Browsing and SSL findings.
Drawings of Buddy and its three companions
From the demo video: Buddy and its three companions, Toru, Bilge and Hayat.
Garanti BBVA sandbox flow
From the demo video: a tier upgrade sent to the Garanti BBVA sandbox with an HMAC-signed request.