Case studies

AI case studies: real projects in production

AI work that is already running. Only projects where the AI is doing a job today. No slides, no prototypes.

AI platform · United States

LegalMente AI

Multi-tenant AI agent architecture and product redesign for a legal platform with patented technology.

Client: LegalMente AI is a legal platform for entrepreneurs in the United States. It forms LLCs and corporations, drafts and reviews documents, and answers legal questions through Para™, an AI paralegal that lives inside the product, not in a help bubble.
The challenge: How do you put a lawyer's work into a form without the form feeling like a lawyer's work? The platform had to serve many customers without mixing their data, and the interface had to keep a first-time founder inside the flow until the filing was done.
What we did: We built the multi-tenant architecture for the AI agents, with isolation per customer, on AWS. Then we redesigned the UI/UX of the parts of the product where a user either finishes or abandons: company formation, orders, cost centers, template library and document view, so the product reads as one tool, not four.
The result: The interface stops being a form the moment the answer is already inside it. A platform with technology patented before the USPTO, running on an architecture we designed.

WhatsApp AI agent + CRM · El Salvador

A WhatsApp AI agent, a Webflow site and a custom CRM showing the same property catalog, in two languages.

Client: Sunrise Discovery sells land, houses and construction services in El Salvador, mostly to buyers living abroad.
The challenge: The site had to feel like a destination brand, but the catalog behind it never stops moving: price per vara², total area, sale type, availability. Marketing, sales and the AI agent on WhatsApp all needed the same truth.
What we did: We built the bilingual site in Webflow and connected its CMS to a CRM we developed on React, TypeScript and Supabase, synchronized through a webhook endpoint. When a property is created, edited, unpublished or deleted on the site, the change reaches the CRM in seconds; a removed listing is marked out-of-market instead of deleted. The WhatsApp AI agent reads from that same source.
The result: A marketing site that behaves like a product. Marketing works in Webflow, sales works in the CRM, and the AI agent on WhatsApp feeds both, with no spreadsheet in between.

AI contact center · El Salvador

Our contact center platform in production: voice, WhatsApp and social channels in one panel, with AI doing the wrap-up.

Client: Tropigas, a gas distributor, and TMCA, a multimodal transport company founded in 1986 and present in five countries.
The challenge: Calls, WhatsApp and social messages were handled in different places by different people, and nobody could measure who attended what, or how it ended. Tropigas also needed the contact center talking to several ERPs.
What we did: We deployed Halo: phone in the browser, WhatsApp Business, Instagram and Messenger in one queue; live supervision with listen, whisper and take-over; order taking and tickets. The AI works at the close of every interaction: automatic summary, suggested tagging and sentiment. For Tropigas we connected the platform to their ERPs so an order taken on a call lands where it belongs.
The result: TMCA has run on Halo for a year, with managed infrastructure. Halo has been stress-tested with 500 calls at 8 per second and 40 agents without errors. Its best feature is knowing when to hand the conversation to a person.