The job
Ritorica sells business messaging: SMS, WhatsApp, Viber, Telegram and push. The website has to explain that to buyers, let the team edit pages without a developer, and turn visitors into leads.
What I built
- Design in Figma, then turned into HTML, CSS and JavaScript templates.
- Content management on Django CMS, so pages can be edited in the browser. PostgreSQL in production.
- Lead capture: contact forms, demo requests and newsletter sign-up.
- An AI chatbot on the Google Gemini API that answers visitors' questions.
- A coverage catalogue with one page per country, at /messaging/.
I build with Claude Code as a pair programmer. I make the decisions, review every change and own what ships.
After launch
- It is now the company's single main site, and it carries the Arabic and Bengali versions.
- I audited the UX of my own live site and scored it 6.4 out of 10. Then I fixed what the score exposed: dead buttons, leftover placeholder text, and statistics that contradicted each other between pages.
- The site was close to invisible in AI search, so I wrote an SEO and GEO strategy with an eight-week checklist and reported progress through Search Console.
- I added an events section covering the company's past events.
Shipping and running it
- Deployed to a Linux server. Changes go live automatically through GitHub Actions, with a systemd timer as a second route.
- The server only accepts SSH keys, and deploys use their own key.
- I wrote the production runbook: what is configured, how to deploy, how to check it worked.
- I worked with the company's system administrator on server access and DNS.
The bug that taught me the most
For almost three weeks the deploy pipeline showed a green tick on every run. Nothing was actually deploying.
The deploy script didn't stop on errors, so each failing step was skipped quietly and the run still reported success. Underneath, the deploy key was broken, an old firewall step timed out for GitHub's servers, and the restart command couldn't run without a password.
I found it by comparing the commit running on the server with the last one I had pushed. Then I replaced the key, removed the step that timed out, gave the deploy user only the permissions it needs, and made the script fail loudly.
What I'd do differently
Add a check at the end of every deploy that compares the live commit with the pushed one, from day one. It's a few lines, and it would have caught the silent failures on the first run.