Problem
Ambientalia has several products (leyrep.cl, ambientalia.cl) that get the same WhatsApp questions, and it also needs a steady stream of communications. Done by hand, all of that eats team time.
Context
An internal tool built at Greensoft. Orbix is “an agent that orbits” Ambientalia's products: on one side it answers WhatsApps with AI, on the other it automates news generation. I planned and built it solo.
My role
- Planned the whole solution and built the backend (no GUI, it runs server-side).
- Built the WhatsApp AI chatbots (LaravelAI) for leyrep.cl and ambientalia.cl (Greensoft.cl on the way).
- Built the news pipeline: scraping in Python (Scrapling) + a local Ollama model (qwen), wired to Bitrix24 over a webhook.
Solution
- Chatbots that answer each site's WhatsApp questions with AI.
- A cron looks for EPR-law news in the configured sources; if there's none, it pulls environmental news.
- The local model summarizes each item into a draft with hashtags and CTA, ready to review in Bitrix24.
- Local AI with Ollama to keep cost and data inside the server.
Views
Representative mockups · no real data, for confidentiality
WhatsApp chatbot
WhatsApp chatbotOrbix
Scraping pipeline
Scraping pipelineOrbix
Generated draft
Generated draftOrbix
Review in Bitrix24
Review in Bitrix24Orbix
Learnings
- Building reliable scraping was the most tedious part of the project.
- Designing a single agent that covers several products and channels (WhatsApp + communications).
Result
In internal use: it answers WhatsApps for Ambientalia's sites and keeps generating news drafts, ready to review in Bitrix24.