Hi, I’m Huseyin Kaplan, building AI into web products.

TypeScript, React, Node.js and Python. RAG, agents and evals running in production.

See my work
NowShipping atRevotas · Istanbul BuildingTedavio AILLM product for clinics WritingEngineering notesRAG, evals and shipping
Reach outBook a 30-minute call

How I work

Small pull requests, written updates, no surprises

Tech stack

The stack behind what I ship

TypeScriptReactNext.jsTailwindViteZod
Node.jsPythonFastAPIDjangoPostgreSQLSupabaseRedis
Claude APIAI SDKLangGraphMCPPydanticpgvector
DockerPlaywrightGitHub ActionsPytestVS CodeCursorGitHub

What you get

Cited answers, tested flows, deployed and measured

Ships with evals · CI gate before merge

Flexible with time zones

Based in Istanbul, available remotely

Istanbul · UTC+3

Uses

The tools I use every day

Windows 11 · Warp · PowerShell 7

Case studies

Curated work

Production systems I built at Revotas. The code belongs to the company, so each one is rebuilt as an interactive demo with synthetic data.

01Internal product2026

Admin platform

Admin console demo with overview metrics, invoices and ticketsA back-office console written from scratch, with cited AI answers over internal documents

Admin platform

2026

A back-office console I wrote from scratch as its only author. Staff run accounts, contracts, invoices, support and billing from one place.

  • CRM for customers, contacts, invoices and agreements, plus support analytics and billing
  • A password-free bridge into any customer workspace, with a guard against incomplete sessions
  • AI Document Center: upload, chunking, vector search and cited streaming answers
  • Command palette, TR/EN, light and dark themes, role-gated modules
  • React 19
  • TypeScript
  • Vite 6
  • RAG
  • ~28k lines
02Storefront script2026

Search V2

Search overlay demo showing results for a query on a synthetic storeOne script that puts search and autocomplete on any storefront, live on 271 brands

Search V2

2026

A search overlay and autocomplete that attaches to a storefront's own search box with one script, configured per brand without code.

  • One IIFE bundle built with esbuild that creates its own DOM and CSS
  • Per-brand config: selectors, synonyms, redirects, recommended queries, preset and CSS
  • Client-side typo tolerance, Turkish folding, facets with counts and zero-result recovery
  • Search and click events for analytics
  • Vanilla JS
  • esbuild
  • ARIA combobox
  • 10 presets
  • 271 brands live
03Storefront script2025–26

Recommendation engine

Recommendation engine demo cloning a storefront product carouselRecommendations rendered in the store's own card design and carousel

Recommendation engine

2025–26

One config-driven script that copies a real product card from the page, fills it with recommendation data and starts the store's own slider on it.

  • Clone mode for Slick, Owl and Swiper, with an isolated root per widget and a full destroy()
  • Native mode with skeleton loading and keyboard support when cloning is not possible
  • Scenario routing by page type and visitor, with fallbacks
  • View, click and conversion attribution
  • Vanilla JS
  • Slick
  • Owl
  • Swiper
  • ~12k lines core
04Panel builder2025–26

Smart widgets

Deal of the Day builder demo with settings and a storefront previewA no-code builder that compiles a marketer's settings into a storefront script

Smart widgets

2025–26

Deal of the Day V2, built from scratch: a builder that turns settings into a self-contained storefront script, with a live preview.

  • Settings compile into one self-contained script, nothing interpreted at runtime
  • Live preview that updates as the marketer types
  • Eight widget designs in the gallery
  • Work across the rest of the conversion widget suite
  • React
  • Vanilla JS runtime
  • Jest
  • postMessage
05React Native2025–26

Mobile push

Mobile push demo with a campaign composer, delivery pipeline and devicePush notifications that open the right screen from any app state

Mobile push

2025–26

The React Native side of push for the platform's mobile clients, from token registration to deep-link routing.

  • FCM token registration and a listener for refreshed tokens
  • JWT login with one automatic re-login and retry on 401
  • Android 13+ permission check and a notification channel
  • Tap routing for foreground, background and cold start
  • React Native 0.79
  • FCM
  • Notifee
  • Deep links
See more projects

Public references

Systems I build

Each card is a public, runnable reference or shipped work. Drag or use the arrows to browse.

Retrieval

Production RAG & Evals · hybrid retrieval with RRF and abstention

Evaluation

LLM Reliability Lab · deterministic evals as a CI gate

Agents with a human gate

Relay Desk · structured actions, approval and recovery

Runtime benchmarks

Agent Runtime Bench · same policy, two runtimes, one CI gate

Founder product

Tedavio AI · private LLM product for clinics

Product engineering

Revotas · admin platform, conversion widgets, RAG assistant

What I bring

From retrieval contract to shipped UI.

I own the whole path: the data, the model call, the evaluation that keeps it honest and the interface people use.

Retrieval & RAG

pgvector, hybrid ranking with RRF, query rewriting, citations and no-context abstention.

LLM evaluation

Golden sets, failure taxonomies and deterministic reports that gate CI.

Agentic tool calling

Native tool loops, structured actions and deterministic human handoff.

Product interfaces

React and Next.js with clear failure and recovery states, tuned for Core Web Vitals.

Backend & data

Node.js, Django and FastAPI services, typed contracts, multi-tenant PostgreSQL.

In production since 2022

4 engineering roles, 6 public projects and one founder product.

Open to roles · replies by email or call
Book a 30-minute call See all projects

Ask about my work

Grounded answers, with sources.

Pick a question. Every answer is written from this site's pages and cites them, the same contract my RAG projects enforce. No model call runs here.

Production AI features and the product around them: RAG with citations and abstention, LLM evals wired into CI, and agent flows with a human approval gate.

From the desk

Latest writing

Read all posts
Three connected browser diagrams: pointer intent, a prepared destination, and an incoming page; a smooth curve links the steps.Web Performance Smooth, without the waitHow I added smooth scrolling and animated page changes to this static site: Lenis, native browser transitions, early preparation, and a test with 30 images still loading.5 min read · Read article Line drawing of a score axis from 0 to 1 with sixty dots in two groups, an empty band between them, one highlighted dot inside the band at 0.35 and a dashed threshold line at 0.5AI Agents A word or a numberThe same refund decision asked three ways, 540 times: a chat model, the same model held to a strict schema, and a decision model that returns a probability. Accuracy tied. The number found the one ticket I had labelled wrong.6 min read · Read article Line drawing of one core script block between four config cards and four storefront plates, one pair highlightedE-commerce Search A results page on someone else's storefrontOne script that renders search results inside stores we do not host: what moved into a per-store config, how it mounts, and why old and new run side by side.5 min read · Read article

Have a problem worth solving?Let's make it reliable

Start a conversation

Open to full-stack and applied AI roles, remote or hybrid.

Tell me about the role, the team and the problem.