Hi, I'm Jovan

Full-Stack + AI Engineer.

I build AI products end-to-end. Right now I’m shipping Knowtis — collaborative notes with a copilot that searches your knowledge base and the web, remembers what matters, and proposes every edit for you to approve before it touches a note.

Open KnowtisGitHub

Now shippinglive

Knowtis copilot

An agent that works inside your notes.

Agent tools: 9
LLM providers: 3
MCP server: 1

Searches your notes and the web
Proposes edits — you approve each one
Remembers across conversations (opt-in)

Vercel AI SDK · pgvector · NestJS · Postgres

Open knowtis.app Read full case study

The Stack

AI-native engineering, end‑to‑end.

The tools I reach for when shipping production-grade AI products — from inline LLM actions to real-time multiplayer and monorepo tooling.

Core craft

AI Engineering

Agents in production — tool loops, retrieval, evals

apps/api/src/modules/agent/infrastructure/tools/note-mutate.tool-group.ts

proposeUpdateNote: tool({
  description:
    'Propose editing a note. Does NOT edit it.',
  execute: async (input) =>
    proposals.capture(
      await build(userId, input)
    ),
});

Vercel AI SDK
MCP servers
Tool-use loops
Human-in-the-loop
Hybrid retrieval
Evals
Injection guards
Claude
pgvector

How it's built in production

Frontend · client craft

Frontend

Modern, accessible, performant UI

React 19
Next.js 16
TanStack Router/Query
TypeScript
Tailwind CSS 4
GSAP

Edge-rendered WCAG 2.1 AA

Backend · server craft

Backend

Type-safe APIs with clean architecture

NestJS 11
Drizzle ORM
PostgreSQL 16
Redis 7
Hono
Python
Django

Ports & adapters · CASL auth

Real-time · multiplayer

Real-time

Multiplayer collaboration & live data

Yjs CRDT
Hocuspocus
Socket.io
WebRTC
WebSockets

Tooling · DX

Tooling

Production-ready monorepos & DX

Nx
pnpm
Vitest
Vercel
Docker
PostHog

Featured · case study

apps/mcp/src/tools/notes.tools.ts

server.tool(
  'notes.read',
  { id: z.string() },
  async (input) =>
    await store.get(input.id)
);

Knowtis

Collaborative notes with an agentic copilot that searches your knowledge base and the web, and proposes every edit for you to approve.

AI actions: 20
LLM providers: 3
MCP server: 1

Key highlights

Agentic copilot with 9 tools over your notes and the web, streamed over WebSocket
Human-in-the-loop: every note mutation is proposed, never auto-applied
Hybrid retrieval — Postgres FTS + pgvector KNN fused with Reciprocal Rank Fusion
Long-term memory across conversations, behind a layered prompt-injection guard

Open Knowtis Case study Repo

More work

Hi, I'm Jovan.