Ship AI-powered web apps and automation in weeks, not months.
Full-stack products, AI agents, and data pipelines, shipped solo. I architect and oversee the system; AI handles the execution. Senior-engineer output in roughly half the time and cost of an agency.
Production systems, not side projects.
Full-stack products, AI agents, data pipelines, and polished client sites. 50+ projects shipped across Upwork and direct engagements.
Lab Data Pipeline
Manual lab operations replaced with a self-updating cloud pipeline.
- Samples
- 310k+
- Schedule
- Daily cron ingest
- Deploy
- AWS EC2
Competitive Intelligence Agent
An AI agent that runs weeks of market research in minutes.
- Research
- 5–15 min / 20 competitors
- Output
- Auto-synced Notion DB
- Reliability
- Async + retry
Jumbo / InterStar
Full e-commerce and ERP system. Storefront returning soon after a compliance update.
CF Moto Store Center
Full-stack store platform for the official regional CF Moto distributor.
Trip Activity Dashboard
PowerBI report over trucking trip and shipment data with driver, dispatcher, and state-level views.
Outlearn Engagement Analytics
PowerBI engagement dashboard for an EdTech platform, delivered in a 48-hour window. Letter of recommendation from the client.
Claims Operations Warehouse
SQL Server warehouse and stored-procedure toolkit for claims operations, with referential integrity and cascading deletes.
AdventureWorks Sales Story
Tableau Public story over the AdventureWorks 2022 Sales schema, exercising LOD expressions, parameters, and cross-dashboard actions.
The Webflow and custom-frontend track that paid for the rest.
Over a dozen client sites, five-star reviews, repeat business.
- Aktiva Aktuel
- Evan Outreacher
- YourExpertly
- Easy Money University
- Design Fantasy
- Austin Pallet Removal
- CA Private · Karl Gustafs
- RDPOOLS
- BIRDLEAD
- Pizza Spot
Three things, shipped deeply.
I pick range over breadth, but only where each area can stand on real, shipped proof.
Full-stack products
End-to-end apps with authentication, data, admin, and real-time features. Next.js, React, Supabase, Postgres, edge deploys.
AI engineering & automation
AI agents, LLM-integrated workflows, scraping and extraction pipelines, MCP tooling, n8n and Make.
Data engineering & analytics
Cloud data pipelines, scheduled ingestion, SQL modeling, KPI dashboards in Power BI or custom.
Four steps, zero theater.
Fewer meetings, clearer scope, faster delivery. The same build in weeks, not quarters.
- 01
Scope in one call
Thirty minutes. I listen, push back, and write down what we are actually building. You leave with a scope doc, a timeline, and a fixed or staged price.
- 02
Architecture before code
Before a single line is written, data models, auth, and integrations are laid out. No surprises in week three.
- 03
Build with AI, review with eyes
Claude Code drafts. I review, refactor, and own every line for security, performance, and reliability. This is where the speed comes from, not the cheapness.
- 04
Ship, document, hand off
Deploy to Vercel or your stack, write a handoff README your team can live in, and stay available for the first weeks after launch.
Engineer's mindset, builder's output.
I come from an engineering background, MSc-level. Systems, measurements, reproducibility. That mindset runs through everything I ship, whether it is a reinsurance analytics CMS or a POS that has to clear a transaction in under 300ms.
Working with LLMs since 2022. AI-assisted development since the tools were good enough to ship production code with. My role is the architecture and the oversight; Claude Code and Codex handle the execution. That split is what makes the timelines and cost structure work, and it is why an individual operator can now outdeliver a mid-size agency on the same scope.
Currently shipping a full store platform for the regional CF Moto distributor, and contributing as a developer on the Macedonian government E-Invoice integration, testing the API and reporting fixes back upstream on behalf of an accounting office.
Outside client work, I run a self-hosted AI agent on a repurposed Linux laptop, reachable over Tailscale and Telegram. It stays off my main workstation on purpose, both for isolation and for the habit of treating agents like any other piece of infra.
Based in Kumanovo, North Macedonia. Remote-native, async-friendly. I answer fast, over-communicate scope changes, and leave codebases cleaner than I found them.
- Location
- Kumanovo, NMK
- Timezone
- UTC+1
- Focus
- Full-stack · AI · Data
- Languages
- English, Macedonian
- Availability
- Q2 2026
- Engagements
- Hourly or fixed-scope
The receipts.
“Filip is both fast and exceptionally professional in his communication. I've worked with 20+ developers and Filip stands out as the best experience I've ever had. Will for sure continue to hire.”
The right projects are specific enough to ship.
A good first call should answer fit, scope, risk, timeline, and the next concrete step.
Good fit
- You have a manual workflow that costs hours every week.
- You need a production web app, admin panel, POS, dashboard, or internal tool.
- Your data lives across spreadsheets, APIs, Notion, Supabase, or Postgres.
- You want one engineer to own scope, build, deployment, and handoff.
Not a fit
- You only need a quick landing page with no product logic.
- You want an AI chatbot without a clear workflow or business outcome.
- You need a large committee, daily status calls, or agency-style account layers.
- You are not ready to share the current process, tools, or constraints.
Workflow audit
1 to 2 daysMap the current process, identify automation candidates, and leave with a build plan.
Prototype sprint
1 to 2 weeksValidate the riskiest part first: AI output, API integration, data model, or internal UI.
Production build
4 to 12 weeksShip the full app or workflow with auth, data, monitoring, deployment, and handoff.
Questions clients ask before we build.
Short answers on AI agents, automation, full-stack apps, data pipelines, scope, cost, and delivery.
Are you writing the code yourself, or is AI doing it?
Both, deliberately. I architect the system, set the constraints, and review every change. Claude Code and Codex handle most of the typing under that direction. That separation is where the speed comes from. I still write the load-bearing parts by hand: data models, security boundaries, anything where nuance matters more than throughput.
How long have you been working with LLMs?
Since 2022, four years. Started with GPT-3.5 wrappers, moved through Claude 2 and 3, then into agentic work, MCP, multi-tool orchestration, and self-hosted runtimes. The NEXUS rig in the Lab is one of those, a personal agent fleet I own end to end.
Why is this faster and cheaper than an agency?
One head from discovery to deploy means no handoff overhead. AI-assisted execution under senior-engineer oversight compresses the implementation half. Together that lands at roughly half the timeline and cost of an agency on the same scope. The trade-off is bandwidth: I take a small number of engagements at once, not a full pipeline.
Do I need a custom AI agent, or is simpler automation enough?
Probably simpler than you think. Fixed-step workflows fit a script or n8n flow. Tasks that need judgment, research, or messy unstructured data justify an agent. Common candidates either way: lead enrichment, document extraction, reporting, support triage, spreadsheet cleanup. The first call filters what shape the build should take.
How long does a build usually take?
A narrow prototype lands in days. A production workflow in a few weeks. A full-stack product with auth, admin, payments, AI features, and deployment is usually 6 to 10 weeks with AI-assisted delivery, against 12 to 16 in a traditional team. Estimates harden after the first scoping call.
How do you keep AI agents safe with private business data?
I treat an agent like a junior operator with limited permissions. It gets only the tools and data it needs, scoped credentials, audit logs on important actions, human approval where risk is high, and fallback paths when model confidence is low.
Can you fix a messy spreadsheet or manual reporting process?
Yes. Manual reporting often becomes a scheduled Python pipeline, a database-backed dashboard, or an internal admin surface. The right shape depends on data volume, who owns the process, and how often the report has to refresh.
Why hire a solo AI engineer instead of an agency?
Speed and accountability. Discovery, architecture, implementation, and QA stay in one head. Fewer meetings, tighter scope, direct ownership of the shipped system. With AI-assisted execution under senior oversight, the output is comparable to a small team at a fraction of the cost.
The fastest way to price a build is to bring one workflow, the current manual steps, and the tools it touches.
Scope a buildReady when you are · 8–12h response
Have a build in mind? Let’s talk.
A thirty-minute call. Bring a rough spec or a vague idea. You leave with a concrete plan and a number.