

Mexico · B2B quote tool · Talk at AI Startup Rodeo, Edition 1, Austin, Texas
TAQuotes: AI to build without being a programmer
- Client and role
- Turbo Air México. We built the MVP, with its features, to test the idea and set the scope. A programmer built the backend and the application that is live today.
- The challenge
- A sales team with more than 2,000 products and no programmers needed to search for products, build quotes with discounts and send them by e-mail. Its basic quote tool was not enough, and nothing on the market covered the way it worked.
- What we did
We built TAQuotes, a B2B quoting platform for the sales team and the distributors of Turbo Air México. It replaces the basic quote tool. The team searches for products, builds quotes with discounts and sends them by e-mail as PDF or Excel, with approval and user roles.
MVP sections:
- Catalog: products, custom products, accessories, spare parts and inventory.
- Cart, with a discount per product.
- Quotes in PDF and Excel.
- Customers.
- Projects.
- Favorites and notifications.
- Admin panel: users, approvals, indicators and analytics.
With the same method we also built the system that translates spec sheets into Spanish on local hardware, and the sales dashboards.
- How
- The sales manager told us the problems of the day-to-day work and what would be nice to have.
- We understand the problem and explain it to the AI.
- We ask for options, not a solution. We research them on our own and keep two or three.
- We go back to the AI with that context, then test, make mistakes and adjust. That is how we built the MVP in several versions, until it was clear what it had to do.
- We review everything. The person is the quality layer, and the project memory lives in files that we maintain.
- A programmer built the backend and the final application on that scope.
- Tools
- We started with ChatGPT, continued with ChatGPT and DeepSeek, and today we work mostly with Claude. To read the PDFs we used a local vision model, LLaVA 13B with Ollama, on a 16 GB graphics card. To give the project a memory we built DeBra, a tool with semantic search and impact analysis. We also used a rules file, correction files by topic, checks before every commit and custom Claude Code skills.
- The talk
- Andrés García, founder of UP & AHEAD Labs: I gave the talk at AI Startup Rodeo, Edition 1, in Austin, Texas, on March 10, 2026, on FullStack AI day. Title: “How Vibe Coding Helped Me Build What The Sales Team Actually Needed”. I showed how a person with no programming training uses AI to build real software. I told the full path, and what failed too. My previous experience was one year of C++ in 2012, a site built during the pandemic and WordPress.
- Result
- Live at taquotes.com.
- 41screens in the MVP
- 1,180products in the catalog
- 4user roles
- 8 monthsof work, from the first prototypes to the current version
- 894commits in 6 repositories
- 300+errors logged in DeBra
- Services



















