We understand your problem first.
Then we solve it with AI.

UNIMINUTO · MIA, the student mentoring agent

Problem

One of Latin America's largest universities could not give continuous, personal guidance to tens of thousands of students with its advisory staff.

What we built

MIA attends students on web, WhatsApp and Teams, detects academic, personal and administrative needs, books tutoring sessions and raises alerts to the Wellbeing team, whose feedback MIA learns from.

Result

20,623 students since June 2024 · 92.8% of conversations rated successful in the last 30 days · 284 tutoring sessions booked

UNIMINUTO · AVA, the admissions agent

Problem

Prospective students arrive by WhatsApp and web at all hours; admissions teams lose them between the first question and the enrollment.

What we built

AVA attends every prospect, handles objections, follows up and hands over to a human advisor when needed.

Result

390 enrollments attributed to AVA in 2026 (figure under reconciliation with the university, which estimates up to 480) · 19,246 WhatsApp conversations with prospective students · 338,485 messages across WhatsApp and web

UNICAFAM · A virtual admissions advisor on WhatsApp

Problem

Between 1,200 and 1,800 new contacts reach the admissions CRM every week, and almost no prospect completes the pre-registration alone: an advisor fills it in by phone, in an academic system with no API.

What we built

A WhatsApp admissions advisor for the 2027-1 campaign (13 undergraduate and 5 graduate programs), traceability of every contact in HubSpot, a web panel with indicators and conversations, document intake by WhatsApp and assisted registration in SIGA.

Result

13 undergraduate and 5 graduate programs for the 2027-1 campaign · between 1,200 and 1,800 new contacts a week · advisor in production since 17 October 2026

Alianza Colombo Francesa · Galia, a French conversation partner from A1 to C1

Problem

Students practise spoken French only in class, and teachers cannot give each one individual conversation time and correction between sessions.

What we built

Galia, an installable web app with avatar, voice and chat integrated into the Alianza's virtual classroom: an autonomous route with curriculum and automatic assessment, a teacher mode that turns a PDF into activities with a rubric, trophies by level and a weekly report.

Result

In production since January 2025, now in its second year with the Alianza · spoken practice from A1 to C1 with automatic assessment · voice response time cut from about 22 s to about 3 s in 2026

UNIMINUTO · Leo, the Socratic tutor inside Moodle

Problem

Students in distance programs study alone between classes, and faculty cannot answer every question in the virtual classroom.

What we built

Leo, an AI tutor inside Moodle that co-constructs answers with the student instead of giving them away. Professors design and publish Moodle activities from Leo, and WhatsApp reminders reach students with pending deliveries.

Result

15,565 students and 1,385 professors in 2026 · 58,730 conversations · 52% of students came back

UNIMINUTO · TOM, the faculty companion

Problem

Hundreds of professors need to align with institutional mission, tools and virtual classrooms, and the induction program cannot reach each of them personally.

What we built

TOM, an AI companion for faculty culture and development that mentors professors and tracks KPIs for the academic leadership.

Result

600 professors invited to the pilot · 551 have logged in and talked with TOM

UNIMINUTO Centro Progresa · HERA, the entrepreneurship mentor across 13 campuses

Problem

Each campus entrepreneurship centre serves hundreds of entrepreneurs with a few mentors. The first diagnostic interview takes 45 to 60 minutes per person, and nobody knows what happens between appointments.

What we built

HERA runs the diagnostic interview by chat or voice, places each business on the institutional matrix (idea, validation, productive unit, company), agrees a real action in the market, follows up, and hands the prepared case to the campus mentor, who decides in 5 to 10 minutes.

Result

2,718 people have talked with HERA · 13 campuses, 26 mentors and 198 entrepreneurs assigned to a mentor at Centro Progresa

Seconpa · LiDAR processing at scale, from solar farms to the Bridge of the Americas

Problem

A surveying company in Panama needed to turn drone LiDAR flights into deliverables (terrain models, contours, classified clouds, CAD drawings) for many projects in parallel, without a processing team of its own.

What we built

The LiDARit processing platform under an annual enterprise plan: Seconpa flies, LiDARit processes and delivers. Projects range from 2-hectare ports to a 310-hectare solar farm, plus the digitization of the Bridge of the Americas to detect cracks and maintenance needs.

Result

19,610 hectares and 12 projects processed by September 30, 2025 · 5 more projects in 2026

SQM Salar · Stockpile measurement in the Atacama salt flat

"We spent 2 weeks in stockpiles measurement using a terrestrial laser, then tried photogrammetry, but it didn't reach the precision we need. LiDARit provides the technology that allows us to get precise and quality results in 3 days."

Problem

Measuring mineral stockpiles with terrestrial lasers took two weeks per cycle, and photogrammetry could not reach the required precision.

What we built

Drone-mounted LiDAR systems (two Miniranger units with RGB photogrammetry tools) delivered to the potassium sulfate plant in the Salar de Atacama, with LiDARit processing.

Result

Stockpile measurement cycle from 2 weeks to 3 days · systems purchased June 2018

Agrodinco · LiDAR for drainage and irrigation design in cacao plantations

Problem

Designing drainage and irrigation for a plantation requires centimeter-level terrain elevation. Photogrammetry could not deliver the vertical accuracy.

What we built

Capture, processing and analysis of high-density LiDAR and RGB data for the Colombiana de Cacao project, delivered as terrain models the engineers could design on.

Result

2,356 hectares delivered between November 2016 and March 2017

WD360 · Construction progress for Mexico's large homebuilders

Problem

Homebuilders need to know, every few weeks, whether each housing development is being built as planned. Walking the site does not scale across cities.

What we built

A LiDAR capture and processing service for WD360, which reports progress to builders such as RUBA and Casas ARA on developments in Querétaro and Jalisco.

Result

Continuous processing contract since November 2024 · monthly operating reviews · 3 developments delivered in 2026

Jolly FinanceOps · The complete accounting operation run by an agent, with human approval

Problem

A small company's accounting arrives as e-mails, electronic invoices and bank statements. Posting, reconciling, DIAN support documents and tax filings all depend on one person's time.

What we built

Jolly FinanceOps: an agent that takes instructions by chat and documents by e-mail, posts and reconciles movements, prepares electronic documents, tax formats and traceable reports. The accountant reviews and approves; nothing is filed or paid without a human.

Result

745 accounting movements from 2020–2026 consolidated and reconciled for a shareholders' review (June 2026) · 10 tax-information formats reviewed and 27 of 27 checks passed on two pre-validated DIAN files (2 June 2026) · 12 billing e-mails triaged to an empty inbox on 23 October 2026

Centro Comercial Único · IAContable, the agent that posts invoices

Problem

The accounting team of a shopping-centre group with four companies spent its days keying supplier invoices and warehouse entries into the ERP, and volume grew faster than headcount.

What we built

IAContable receives invoices through Microsoft Teams, reads them, posts them into the Siesa ERP three times a day and learns from each correction the accountants make.

Result

Live since April 2025 · 2,286 invoices read and 326 posted autonomously in the first months · still operating in October 2026

Industrial fan manufacturer (Cali) · A consultative sales agent on WhatsApp

Problem

WhatsApp attention was inconsistent: the advisor asked again for data she already had, offered the wrong equipment and dropped the conversation after sending the datasheet. The criteria for choosing the equipment were not written down.

What we built

A consultative sales agent on WhatsApp that qualifies the lead, sizes the equipment, quotes, sends datasheets and images and follows up with Meta-approved templates, plus a web CRM with a 14-stage pipeline, every conversation, AI or manual mode, reports and forecasts.

Result

Monthly sales of the commercial line two to three times higher in September 2026 than before the agent, all originated in the CRM (client-reported) · the agent handles 487 of 610 active leads (80%) as of 20 October 2026 · in production since 11 June 2026

Restaurant and supermarket operators · NOMI, the AI host

Problem

Restaurant and supermarket operators lose customers to slow attention and lose staff to weak onboarding.

What we built

NOMI, an AI host that attends customers and trains new staff on the operator's own menus and procedures.

Result

Deployed for a Panamanian operator's supermarkets and restaurants under contracts signed in November 2024 and January 2025