LLM features in production
RAG, structured output, streaming, and prompt orchestration — integrated into your React/Node product and built to survive real traffic.
Milan Milenkovic
I'm an ML & Applied AI Engineer and full-stack developer. I build production ML pipelines, integrate LLMs, develop agentic systems, and ship RAG pipelines into live products — alongside the search, auth, and infrastructure they run on. I work primarily as a fractional / part-time Applied AI engineer (~10–15h/week) for EU and US teams that want production AI features shipped steadily, without hiring a full-time headcount.
What I do
ML & Applied AI engineering, end to end
LLM integration, agentic systems, and RAG in React/Node. Full-stack is how I deliver — not a separate service line.
Agentic systems
AI agents with tool use, function calling, and autonomous workflows that take real actions inside your product — not just generate text.
Full-stack delivery
React, Next.js, Node.js, vector DBs, deployment — I own the stack end to end so your AI features actually ship, not stall in a notebook.
ML & Applied AI Engineering
How I build with AI
From retrieval and structured LLM output to agents that call tools and automate workflows — engineered for production in the JavaScript stack.
RAG systems
Ground LLMs in your data with embeddings, vector search, and retrieval tuned for your domain.
LLM app development
Structured output, streaming, prompt orchestration, and safe integrations that survive production traffic.
Agentic code & automation
Agents with tool use, function calling, and multi-step reasoning that automate real workflows inside your codebase and product.
Evals & observability
Testing, evaluation, and monitoring so AI features stay accurate after launch.
Work
Shipped & live
Real products with real users — numbers included, not just adjectives. AI-led first: najdiavto.com ships a natural-language search feature on top of a production marketplace.
najdiavto.com
LiveAI search live
200+car brands
212municipalities
AINL search
Production car marketplace with real dealer inventory — faceted search plus natural-language AI search across fuel, transmission, body type, color, doors, and Euro norm.
View case studykembo.app
LiveProduct / infraFree tier · Pro plans
4platform SDKs
OAuthhosted flow
Prodmode ready
Hosted auth infrastructure for mobile and web — OAuth, short-lived JWTs with rotating refresh tokens, hashed sessions, role-based user management.
Visit sitesignpad.online
LiveProduct / infra
0installs needed
100%in-browser
Freeto use
Browser PDF workspace — upload, edit, annotate, and sign documents with no install, no ads, and a proper landing page.
Visit siteLabs
ML Projects & experiments
Open-source ML projects that demonstrate specific techniques — from NLP to recommendation systems.
najdiavto-ml
LiveCase study
0.86MRR (LTR)
62eval queries
4MLOps phases
End-to-end ML service for a car marketplace — price intelligence, hybrid search, learning to rank, and MLOps with monitoring and safe retraining.
View case studycomment-sense
LiveCase studyFeedback loop
3languages
6categories
HybridML + lexicon
Multilingual NLP demo — sentiment analysis and text classification in English, Serbian, and German with a feedback loop for continuous improvement.
View case studymatchmaker
LiveCase study
9,700+movies
HybridSVD + content
3languages
Hybrid movie recommendation engine — collaborative filtering (SVD) + content-based genre similarity. Powered by MovieLens with 9,700+ films.
View case studyautoml-lite
LiveCase study
3models
Autofeature detection
R²best model
Automated ML pipeline — upload data, train regression models (LinearRegression, Ridge, RandomForest), compare metrics, and make predictions. No code needed.
View case studyHow I work
A simple path to a fractional AI engagement
You work with me directly — no sales team, no account manager. I scope honestly, ship weekly, and care about what happens after launch.
Find the highest-leverage AI move
A focused kickoff: review your data and product, pin down the highest-leverage AI feature, and agree on metrics that define success.
Ship a working slice fast
Within the first cycle I deliver something real in production — a RAG endpoint, a structured LLM feature, or an automated workflow — not a slide deck.
Then keep the momentum
We decide together: a steady fractional retainer (~10–15h/week) to keep shipping, or a scoped fixed project. No long-term lock-in.
What you get
- Short feedback loops — you see working software every week
- Direct communication, no account-manager layer
- Production-first: metrics, latency, and reliability matter
- Clean, maintainable code you can hand off
- Flexible engagements — fractional retainer or fixed project
Need an Applied AI engineer on a fractional basis?
Tell me what you're building — I'll get back within 24 hours. No obligations, fully confidential. Open to fractional / part-time retainers (~10–15h/week) across the EU and US.