I start with the business problem and pick the technology after. My work at Webvoltz covers three areas:

• Automation that takes repetitive tasks off your team, safely and without mistakes.
• AI over your own data, with answers that show their sources instead of guessing.
• Real-time systems that stay fast and accurate when many people use them at once.

Every project is backed by automated tests, so it keeps working after launch.

The challenge: Ticket buyers in a large venue need to pick exact seats, and two people must never end up with the same seat.

What we delivered:

• A live seat map for a 15,000-seat venue that responds instantly on every click.
• Only the section a buyer opens is loaded in detail (1,500 seats at a time), which keeps it fast: the overview is just 66 shapes, and the whole map is a 75.8 kB download.
• Seat holds expire on time, every buyer sees other buyers' holds live, and clashes resolve the same way for everyone. A seat is never sold twice.
• Fully accessible: works with screen readers and keyboard, plus a "best seats for my group" finder. Backed by 222 automated tests.

_{Built with React, TypeScript, Redis, PostgreSQL, WebSockets}

The challenge: A reputation agency serving local clinics and service businesses had front-desk staff asking customers for reviews by hand.

What we delivered:

• Staff approve a happy customer with one click, and the system runs the full review-request campaign inside the client's GoHighLevel account.
• 8 safety checks before any text goes out: waiting period, allowed sending hours, opt-outs, customers who already reviewed, and frequency limits.
• Every reply, click, opt-out and failure is tracked, and each one is processed exactly once, so no customer is ever texted twice.
• A second integration sends new prospects straight into the agency's sales pipeline without creating duplicates. 11 automated triggers, 141 tests.

_{Built with Node.js, TypeScript, PostgreSQL, GoHighLevel API, React}

The challenge: Teams lose hours searching long contracts, reports and manuals, and most AI tools give answers you can't verify.

What we delivered:

• Upload a document, ask a question in plain words, and get an answer where every fact points to its page.
• It keeps the document's chapters and sections intact, so answers never lose their context.
• Simple questions get a quick answer. Complex ones get a research agent that reads the right sections. "List every deadline" reads the whole document.
• A second AI checks every sentence against the source and removes anything the document doesn't support.

_{Built with Python, FastAPI, PostgreSQL (pgvector), Claude}

The challenge: Most AI products that search the web depend on someone else's search engine, so they can't control results, freshness or cost.

What we delivered:

• An independent AI search engine with no third-party search API. Finding, reading, indexing and ranking pages are all in-house.
• Smart crawling that revisits fast-changing pages often and leaves static ones alone.
• Search by keyword and by meaning combined, with an explanation of why each result ranks where it does.
• AI answers that can't cite pages they didn't read.

_{Built with Python, Scrapy, OpenSearch, PostgreSQL, FastAPI}

The challenge: CCTV networks have far more cameras than people to watch them.

What we delivered:

• A proof of concept that finds and tracks every person and every gun in live CCTV footage.
• A second, zoomed-in look around each person gives distant weapons about 2.2× more detail, so small guns aren't missed.
• Analyzes each frame in 35–95 milliseconds on a single GPU, with every alert drawn on exactly the right frame.

_{Built with Python, YOLO, ByteTrack, PyTorch, FastAPI}

The challenge: Insurance, property, logistics and field teams need to make decisions based on location.

What we delivered:

• A practical guide to map features: drawing zones, colour-coding data, address search, "which zone is this in" lookups and distance tools.
• Based on a real reference project backed by 1,350+ automated tests.

_{Built with Mapbox GL, Turf.js, PostGIS}

• Problem first, technology second. I find the step that costs your team the most time or money, and fix that first.
• No silent failures. Retries, safety checks and exactly-once processing, so nothing is lost and nothing happens twice.
• AI you can trust. Answers come with sources, and anything unsupported is removed.
• Tested against the real thing. Automated tests run on real databases and services, not just mock-ups.

Area	Tools
Web front end	React · TypeScript · Tailwind CSS · Accessibility
Back end & APIs	Node.js · Express · Python · FastAPI · WebSockets · Webhooks
Databases & search	PostgreSQL · pgvector · PostGIS · Redis · OpenSearch
AI	RAG · Hybrid search · AI agents · Claude · Computer vision (YOLO, PyTorch)
Integrations	GoHighLevel · Mapbox
Delivery	Docker · GitHub Actions · GitLab CI · Automated testing