$ _

AI Automation & Integration Developer

I turn messy manual work into AI-powered pipelines and full-stack apps that ship.

Freelance developer based in Bulgaria. I build production-grade automations on n8n, OpenAI & Anthropic APIs, and LangChain, backed by full-stack engineering in Python/Django/FastAPI/React and .NET/Blazor.

View Projects Get in Touch Available for new projects
GitHub Upwork Email
6+Shipped projects
2LLM providers integrated
5Production n8n workflows
BGBased in Bulgaria, remote-first

01 · About

Engineer first, automation specialist by trade.

I'm Velimir, a freelance developer who builds the layer between "we have a manual process that eats hours every week" and "it just runs itself now." That usually means wiring together LLM APIs (OpenAI, Anthropic), orchestration tools like n8n, and a properly engineered backend so the automation doesn't fall over the first time it sees a malformed input.

Before automation became the focus, I built full-stack applications (POS and inventory systems, admin dashboards, internal tools) with Python/Django/FastAPI and React on the frontend, plus .NET/Blazor for a different flavor of full-stack. That background is why the automations I ship come with real API layers, schema validation, retries, tests, and CI, not just a fragile chain of no-code nodes.

  • FocusAI automation, LLM integration, full-stack web apps
  • Core toolsn8n · OpenAI API · Anthropic API · LangChain
  • Also build inPython/FastAPI/Django · React · .NET/Blazor
  • AvailabilityOpen to freelance & contract work

02 · Skills

What I work with.

AI & Automation

OpenAI APIAnthropic Claude APILangChain n8nPrompt EngineeringStructured Output / Pydantic RAGLLM Eval Harnesses

Backend

PythonFastAPIDjangoSQLAlchemy .NETBlazorPostgreSQLSQLite

Frontend

ReactJavaScriptJinja2 HTML5CSS3Responsive UI

DevOps & Quality

Docker & Docker ComposeGitHub Actions CI pytestGitREST API DesignRBAC / Auth

03 · Work

Selected projects.

A few things I've built end-to-end: from database schema to the API layer to the UI.

AI Automation · RAG & LLM Tooling

RAG Shop Assistant

A hybrid-retrieval shopping assistant for clothing e-commerce: an embeddable chat widget backed by a Claude tool-use loop over a Postgres + pgvector catalog, with a stylist layer (occasion, formality, colour-pairing) on top of plain product search.

  • FastAPI agent service running a Claude tool-use loop over 5 retrieval tools plus a dedicated build_outfit tool
  • Multi-tenant Postgres + pgvector schema; a new shop's catalog and knowledge base import via config file, no code changes
  • Stylist layer: occasion/formality/colour-pairing product tagging and gender-aware outfit templates across 15 occasions
  • Production hardening for shared multi-tenant SaaS: per-shop origin allow-listing, Redis rate limiting, usage metering, monthly conversation caps, bring-your-own API key, prompt caching
  • Embeddable, Shadow-DOM-isolated JS widget (single script tag), an eval harness, and CI
PythonFastAPIAnthropic Claude API PostgreSQLpgvectorRedisJavaScriptDocker
Book a Demo See Pricing

AI Agents · QA Tooling

Manual QA Agent

A Claude-powered manual QA tester for web apps: it launches a real Chromium browser, writes its own test plan, then clicks/types/navigates through the app like a human tester - checking that flows complete, forms validate input properly, and the layout stays stable - and documents everything to an .xlsx report.

  • Monorepo with one shared engine (Playwright + Claude agent loop + report generation) powering a CLI, a Windows desktop app, and a browser-based web app
  • Two modes: full exploratory QA (agent builds and runs its own 5–15 case test plan across the whole app) or guided scenario QA (plain-English scenarios only)
  • UI-quality checklist on everything it touches: probes form fields with invalid input to confirm real validation feedback, and diffs screenshots before/after each action to catch layout shifts
  • Black-box guarantee: no file-reading tool, never inspects source code - every test case comes from what's actually rendered, exactly like a human tester
  • Multi-tab .xlsx report (Summary, Test Plan, Test Results, Findings, Action Log) plus a full Playwright trace for step-by-step replay
Node.jsPlaywrightAnthropic Claude API ElectronExpressxlsxMonorepo
Book a Demo See Pricing

Automation · Local Business SaaS

Repute

An automated Google review request system for local businesses: sends a 3-message SMS sequence (check-in → review request → reminder) after each completed job, tracks review-link clicks, and follows up automatically if the customer hasn't clicked.

  • n8n-orchestrated 3-step SMS sequence over Twilio, driven by scheduled Supabase queries, with click tracking and auto-redirect via a small FastAPI service
  • Reactivation campaign script imports a CSV of past customers and drops them straight into the review-request step, in throttled batches
  • AI-drafted review-reply generator that logs drafts for approval and never auto-posts; Google Business Profile polling is stubbed pending API approval
  • Multi-tenant groundwork: a clients table so the reactivation and review-response scripts work by client ID instead of hardcoded values
  • Runs fully locally with no external accounts (SQLite + mock SMS/draft backends), full pytest suite, swappable to Supabase/Twilio/OpenAI via env vars
PythonFastAPIn8nSupabase TwilioPostgreSQLSQLitepytest
Book a Demo See Pricing

AI Automation · Backend Architecture

AI Intake Triage Pipeline

A reusable pipeline that watches an intake source, uses an LLM to classify, score, and extract structured data from unstructured text, then routes the result to Slack, a generated draft, or a human review queue. Demoed with job-posting intake, but built to generalize to lead qualification, support triage, and RFP screening.

  • Pluggable LLM layer: same schema whether it's running on OpenAI or Anthropic
  • n8n handles orchestration/integrations; FastAPI does structured parsing, retries & fallback
  • Postgres audit log doubles as a labeled dataset for tuning prompts and thresholds
  • pytest suite + GitHub Actions CI, plus a standalone eval harness for the LLM layer
  • Live dashboard to submit and watch items get scored, categorized, and routed in real time
PythonFastAPIn8nOpenAI AnthropicPostgreSQLDocker Composepytest
View on GitHub →

Full-Stack · Business Systems

Clothing Store Inventory & POS

A local web app running a real clothing store's full operation (inventory, purchasing, point-of-sale, and reporting), built server-rendered with FastAPI and Jinja2, no JS framework or build step required to run it.

  • Size/colour product variants, per-variant SKUs, unlimited nested categories
  • Purchase orders track each delivery as its own stock batch; profit computed FIFO per sale
  • Cart-based POS checkout with search/category browsing and per-sale receipts
  • Revenue, cost, margin and top-seller reporting with date-range filtering
  • Session-based auth with admin / sales roles; bilingual UI (Bulgarian + English)
PythonFastAPISQLAlchemySQLite Jinja2RBACi18n
View on GitHub →

n8n · Productized Automation

n8n AI Workflow Templates

A library of ready-to-import n8n workflows that pair GPT-4o mini with real business automations, each one a single JSON file, importable into any n8n instance in seconds.

  • AI email triage: classifies & routes inbound email, alerts Slack, logs to Sheets
  • Daily AI news digest: pulls RSS feeds, summarizes with AI, emails an HTML digest
  • AI lead scoring: scores inbound leads 1–10 and flags hot leads to sales in real time
  • AI invoice extractor: turns raw invoice text into structured JSON for ERP/accounting
  • AI content repurposer: one article in, five platform-ready formats out
n8nOpenAI (GPT-4o mini)Slack API Google Sheets APIGmail APIWebhooks
View on GitHub →

04 · Process

How I approach a build.

01

Understand the process

Map the manual workflow as it actually happens today (inputs, decision points, exceptions) before writing a line of code.

02

Architect for reality

Design the schema, routing logic, and failure paths first. Malformed input and provider outages are expected, not edge cases.

03

Ship with tests

Automations ship with a test suite, CI, and an eval harness to catch quality regressions where an LLM is in the loop.

04

Hand off & support

Clear docs, a Docker Compose setup that runs in one command, and support adapting the system as the client's process evolves.

05 · Pricing

Simple, transparent pricing.

Tiered pricing for productized offerings like Repute and RAG Shop Assistant is coming soon. In the meantime, reach out and I'll put together a quote based on your specific project.

Coming soon

Starter

For a single business getting the core automation running.

TBA
Get notified

Coming soon

Growth

For teams that want the full workflow, integrations, and support included.

TBA
Get notified

Coming soon

Enterprise

Custom volume, multi-tenant, or white-label needs.

TBA
Contact sales

06 · Contact

Have a process worth automating?

I'm currently taking on freelance and contract work: AI automation pipelines, LLM integrations, or full-stack builds. Send a message or request a demo of an existing product, and I'll get back to you within a day or two.

velimir.hr.r@gmail.com Hire me on Upwork GitHub Profile

Typically replies within 24–48 hours

Please enter your name.

Please enter a valid email address.

Please choose an option.

Please add a short message.