▄▄▄██▀▀▀█ ██ ██▓ ██▓ ▄████▄ █████
▒██▒ ██ ▓██▒▓██▒ ▓██▒▒██▀ ▀█ ▒██▓ ██▒
░██░ ▓██ ▒██░▒██░ ▒██▒▒▓█ ▄░▒██▒ ██░
▓██▄██▓ ▓▓█ ░██░▒██░ ░██░▒▓▓▄ ▄██░░██ █▀ ░
▓███▒ ▒▒█████▓ ░██████▒░██░▒ ▓███▀ ░░▒███▒█▄
▒▓▒▒░ ░▒▓▒ ▒ ▒ ░ ▒░▓ ░░▓ ░ ░▒ ▒ ░░░ ▒▒░ ▒
▒ ░▒░ ░░▒░ ░ ░ ░ ░ ▒ ░ ▒ ░ ░ ▒ ░ ▒░ ░
░ ░ ░ ░░░ ░ ░ ░ ░ ▒ ░░ ░ ░
░ ░ ░ ░ ░ ░ ░ ░ ░
░
> Connection established. Welcome to julicq@github:~$
julicq@github:~$ cat /etc/motd ┌──────────────────────────────────────────────────────────────┐
│ Julian Pavlov · Builder of AI products & full-stack systems │
│ Location: Novi Sad, RS / GMT+2 │
│ Shell: zsh + Warp │
│ Editor: Cursor · Xcode · vim (when nostalgic) │
└──────────────────────────────────────────────────────────────┘
julicq@github:~$ whoami --verboseidentity:
role: "AI engineer / founder"
focus: ["agentic systems", "LLM apps", "RAG pipelines", "iOS apps"]
taste: "ships > talks. Pragmatic over fancy."
mission: "Build agents that do real work — not demos."julicq@github:~$ curl -s https://pijeon.dev | less ┌──────────────────────────────────────────────────────────────┐
│ πJeon Dev · Software, with intent. │
│ The engineering house I founded · Novi Sad & Remote │
└──────────────────────────────────────────────────────────────┘
Machine learning, software engineering and design as one craft
with three instruments. Less wins when it does the same job.
INTELLIGENCE autonomous agents, RAG + retrieval evals,
fine-tuning on private data, forecasting
ENGINEERING typed, tested, instrumented backends · APIs,
infra, observability, data pipelines
DESIGN identity, interfaces, information architecture
PRODUCTS
Relay control plane for an agent you already run:
keys, a pause switch, citations from your files
and a log that lasts. HTTP / MCP, no chat.
Autopilot task in, checked Change out. The model writes
the code; tests, an independent review and a
security audit decide. Runs inside your network.
Odgovor one line of HTML puts an assistant on a small
business site. Answers from the site, hands off
to a person when the site does not say.
Custos™ independent mark for AI used in schools: a named
product, at a named version, public methodology.
julicq@github:~$ ls -la ./current-projects/drwxr-xr-x Pulsoric # Agentic LibreChat for audit teams · custom agents, matters, SSO
drwxr-xr-x screenplay.tech # FastAPI narrative-structure analyzer for screenwriters
drwxr-xr-x client-day # Pre-sales CRM · FastAPI + SwiftUI/SwiftData iOS client
drwxr-xr-x immersa-lang # Language-learning platform · web + native iOS
drwxr-xr-x moral-cut # Next.js editorial product
-rw-r--r-- more... # plenty more in the oven
julicq@github:~$ cat ./stack.txtLANGUAGES Python · Swift · TypeScript · SQL · Bash
AGENTIC Claude Agent SDK · OpenAI Agents · LangGraph · CrewAI
MCP servers · tool-use loops · multi-agent orchestration
prompt caching · context engineering · evals
AI / ML OpenAI · Anthropic · Gemini · LangChain · RAG · vector DBs
PyTorch · scikit-learn · pandas · numpy
BACKEND FastAPI · SQLAlchemy 2.0 · Alembic · Jinja2 · Pydantic
Flask · Django · slowapi · aiosqlite
FRONTEND Next.js · React · TypeScript · Tailwind · Vite
MOBILE SwiftUI · SwiftData · Combine · TestFlight pipelines
DATABASES PostgreSQL · SQLite · Redis · MongoDB
INFRA Docker · docker compose · systemd · Nginx · Ubuntu
AWS · GCP · Azure · self-hosted VPS
TOOLING Claude Code · Cursor · git · GitHub Actions · Postman
DESIGN Figma · Blender · Photoshop · 3D / motion when it helps
julicq@github:~$ cat ./agentic-philosophy.md# How I build with agents
1. Agents > chatbots. A model that can read, plan, call tools and
verify its own work beats a prettier prompt every time.
2. Tools are the product. The agent is only as useful as the tools
you give it — design them like APIs, not afterthoughts.
3. Context is engineering. Caching, memory, retrieval and pruning
are first-class concerns, not optimizations for "later".
4. Loops, not one-shots. Real work happens in iterate→observe→correct
cycles. Plan for self-recovery, not just happy paths.
5. Evals or it didn't ship. If you can't measure agent quality,
you're shipping vibes.
6. Humans in the loop where it matters. Autonomy is a slider, not
a switch — turn it up where the cost of a wrong action is low.
julicq@github:~$ tail -f ./now.log[INFO ] building agentic workflows for audit, sales and education verticals
[INFO ] wiring MCP servers + custom tools into Pulsoric and internal stacks
[INFO ] shipping iOS apps with SwiftUI + SwiftData
[INFO ] deploying FastAPI services to bare-metal Ubuntu boxes
[INFO ] obsessed with prompt caching, agent loops and tool-use ergonomics
[INFO ] open to interesting AI / agentic / iOS collaborations
julicq@github:~$ contact --listjulicq@github:~$ git log --stats --since="forever"> exit
Connection to julicq@github closed.
> _


