Six ways to bring AI into your company
From deciding where to invest to running the system in production. Every project starts with a pilot on your data, so you see results before investing more.
AI agents
Agents that serve your customers, answer with reliable information and handle requests for them.
Your customers and your team ask the same questions again and again, and the answers are spread across documents, systems and databases. We build AI agents that talk with them, answer only with what your documents and data say, and then act: they create the order, log the request or hand off to a person when needed.
I want an agent for my companyWhat we do
- Agents that answer from your documents, with exact answers and their source, using hybrid search (RAG and BM25).
- Agents that answer questions about your data in plain language, with no SQL: ask your data.
- Action-taking agents that, beyond answering, carry out the task in your systems.
- Semantic search: your customers find your products by describing them in plain language.
- Image search: your customers upload a photo and find the product they are looking for.
- Agents that read, classify and route documents and requests to the right team.
- Agents with Claude, Gemini or other models, on the platform you prefer: the direct API, AWS Bedrock, Google Cloud or Azure.
Cases
- Dropi: an agent that turns questions into SQL queries, then checks and corrects them on its own; CAS-IA, which handles customer requests; semantic and image-based product search; and the foundation of a support agent on AWS Bedrock and Strands.
- Colpensiones: classification and routing of handwritten documents with accuracy above 85%, in production since 2021.
- Reybanpac (Ecuador): an agent that answers questions in Spanish about production and costs on banana farms.
- Predizer Tenders: reviews public tenders across Latin America and shows the ones that fit each company. INORCA and MUBI Medical use it.
- Universidad de San Buenaventura, MAAS Transmilenio and PROQC International: data integration and BI, the foundation for agents that answer about your data.
AI process automation
Long, tedious tasks where mistakes are costly, done on their own and safely.
Many teams lose hours on repetitive tasks where a mistake is expensive. We automate them through the route that best fits your systems: through the web, with agents that operate portals and applications the way a person would, or underneath, directly on the database with stored procedures and PL/SQL. We work inside your infrastructure, with minimal access that you control, and we build with AI and SDD to deliver in weeks.
Automate a processWhat we do
- Web automation: agents that query portals and systems and record the results.
- Database automation: stored procedures and PL/SQL on your databases.
- Data extraction from PDFs, images and handwritten documents.
- Automated checks that catch errors before they cause problems.
Cases
- ADRES: automatic lookup of each member's health insurer (EPS) and health-system status for a compensation fund, delivered in under a week.
- SIMON: management of the paperwork for a compensation fund's certifications, delivered in under a month.
- Document type: automatic detection of typing errors in members' ID document types, delivered in under two months.
- ABC costing: the spreadsheet a company used to calculate its activity-based costs, which only one person understood, turned into an automated, documented system others can use.
- A telecom company with operations across Latin America: ticket management with RPA for its operations in Colombia, Venezuela, Peru and Chile, delivered in under two months.
- Cámara de Comercio de Cali: automatic data extraction from PDF and TIFF files, cutting response time, typing errors and staffing costs.
- INFODEC: item extraction from payment receipts.
Predictive and prescriptive machine learning models
Decide with data: what will happen and why it happened.
How much you will sell or produce, which customer will stop paying, which student will drop out, why a machine failed. We build models that answer those questions, explain each result and include a simulator to test scenarios.
I want to forecast in my companyWhat we do
- Sales and demand forecasts by product, store and month.
- Agricultural and industrial production forecasts.
- Credit risk and customer or student churn.
- Logistics: delivery times, returns and carrier choice.
- Root-cause analysis of failures from sensor data.
Cases
- JGB and Carvajal: unit forecasts by product and store with 98% accuracy.
- Ingenio Mayagüez: sugarcane production forecasts with 97% accuracy.
- Reybanpac: a pilot banana-production forecast with a weighted error (wMAE) below 6%.
- Credifinanciera: first-horizon default-prediction accuracy rose from 55% to 90%, and the credit process needed 90% fewer questions.
- Universidad Javeriana and Universidad Autónoma de Occidente: detection of students at risk of dropping out with 93% accuracy.
- Grupo Vardí: forecasts of which car models will sell, and how many, over the coming months.
- Smurfit Kappa: explainable AI on more than 1,000 sensors sampled every 5 seconds, to understand why the mills failed.
AI-assisted software development and SDD
Complete systems in weeks, not months.
AI makes it possible to write software much faster, but without a clear specification the result is hard to verify and maintain: that is vibecoding, and it is not what we do. We use spec-driven development: first we agree with you on what the system must do and how we will check it, then we build it with coding agents and automated tests.
Tell us what you need to buildWhat we do
- A specification with requirements, acceptance criteria and design, which you approve before work starts.
- Development with coding agents (such as Claude Code) guided by that specification.
- Automated tests that check every criterion.
- Documentation so your team, or your own agents, can maintain and extend the system.
- Coaching so your developers can adopt this way of working.
Cases
- A compensation fund: the three automation systems: ADRES in under a week, SIMON in under a month and document type in under two months.
- CarLove Unicentro: an information system for an auto paint shop (job intake, task assignment and progress tracking), with an AI agent to operate it by text, delivered in under two months.
- Reybanpac: an analytics agent built with Claude Code, with automated tests for every component.
AI strategy and governance
Invest in the AI that actually serves your business.
Everyone talks about AI, but not every use creates value. We help you choose where to invest, estimate the return and manage the risks, drawing on our consultants' more than 30 years in IT management at several companies.
Book an assessmentWhat we do
- Assessment of AI opportunities in your company.
- Selection and prioritization of use cases by expected return.
- Implementation roadmap.
- AI governance model and policies aligned with the EU AI Act, NIST AI RMF and GDPR.
Cases
- Coomeva: AI governance model based on GDPR, NIST AI RMF and the EU AI Act.
AI training
Help your team use AI with good judgment, starting tomorrow.
Hands-on courses and workshops designed for your industry and processes, for executives and technical teams.
Let's design a training for your teamWhat we do
- Courses for senior management: what AI is, how others in your industry use it, opportunities and risks.
- Workshops for teams with tools such as Claude and Gemini applied to their daily work.
- AI-assisted and spec-driven development workshops for software teams.
Who has trained with us
---
Do you have a process that AI could do better?
Tell us what you want to solve. In a first call we will tell you whether AI is the right tool and what a pilot would look like. We reply within one business day.