REACT · FASTAPI · PGVECTOR
Atlas RAG
A retrieval-augmented knowledge assistant indexing 2M+ documents with hybrid search. Sub-400ms answers with inline citations.
ReactFastAPIpgvectorOpenAI
Open live demo →
● Available for hire · 2026 |
Building apps
people actually use.|
AI developer specialising in RAG systems, agents, chatbots and automations — wrapped in fast, accessible interfaces built to scale.
ReactTypeScriptPythonLangGraphPostgres
// building the future
const gjokica = {
role: "AI Developer",
stack: ["React", "Python", "TS"],
builds: "RAG · agents · automations",
open: true
}
// 01
Projects.
Selected · 2024—2026NEXT.JS · AI SDK · SQLITE
PERIPHERAL
A fictional dev-hardware shop with an AI support agent that answers only from the store's catalog and policies — and opens a ticket when it can't.
Next.jsAI SDKTool callingSQLite
Open live demo →
NEXT.JS · AI SDK · EVALS
Groundtruth
An eval lab for the Atlas assistant: 16 frozen cases, deterministic assertions, and a retrieval ablation you can re-run in the browser.
Next.jsAI SDKEvalsStreaming
Open live demo →
// 02
The stack.
FRONTEND
React
Next.js
TypeScript
Tailwind
Motion
AI / BACKEND
Python
LangGraph
pgvector
Node.js
Postgres
// 03
About me.
I'm Gjokica — an AI developer from North Macedonia working with founders and product teams across Europe. I started in full-stack web and moved deep into applied AI when retrieval and agents became genuinely useful rather than a demo.
My approach is pragmatic: define the evaluation first, ship the smallest working system, then harden it. No model worship, no unnecessary abstractions — just systems that hold up under real traffic.
Today I build RAG assistants, tool-calling agents, chatbots, workflow automations and the production web apps that wrap around them.
6 yrs
Experience
40+
Projects shipped
OSS
Contributor
Remote
Friendly
// 04
Open source.
2.4k
STARS EARNED
38
PUBLIC REPOS
1,830
CONTRIBUTIONS · 2026
rag-toolkit 1.2k
Composable retrieval pipelines with evals baked in.
Python
flowbot 740
Type-safe chatbot runtime for tool-calling agents.
TypeScript
vec-bench 318
Benchmark harness for vector store recall and latency.
Rust
// 05
From clients & teams.
"Gjokica shipped our RAG assistant in three weeks. It handled support volume we thought needed a whole team."
Marta Kalinowska
HEAD OF OPS · NORDWAVE
"Rare combination: he thinks like an ML engineer and builds interfaces like a product designer."
Dario Bianchi
CTO · VENDRA
"Precise, fast, and painfully honest about scope. Exactly what you want in a technical partner."
Elena Petrov
FOUNDER · CODA
Let's build something.
Currently available · 2–3 project slots open
Chat answers instantly · email within 24h