Custom software
AI · Data · Automation
Bogotá, Colombia
Est. MMXXVI

We auto­ mate.

We build software, predictive models and AI agents that work for you — and keep running when nobody is watching.

Web and mobile apps that reach production and the app stores. Not prototypes: products people use.

See projects

Models that anticipate prices, demand and risk. Uncertainty gets measured, not invented.

See projects

Assistants that read your data and answer with evidence. And we teach you to build them.

Go to the Academy

We connect your tools and delete the repetitive work. It runs itself, every week.

See projects

From messy CSVs to dashboards people actually decide with. A reliable pipeline behind every number.

See projects

We don’t sell no-code or pretty demos that fall over on the first Monday. Behind every workflow there is engineering judgement: well-modelled data, code that can be maintained and machines that really do run on their own.

— Automatizaciones Digitales S.A.S. Edwin García · Engineer & founder
[ What we do ]

What we put
to work

02

AI & predictive models

Models that forecast prices, demand and risk — with uncertainty quantified, not invented.

→ 26 years of market data behind the models

PythonLightGBMMarkovConformal
03

AI agents & copilots

Assistants that read your data and answer with evidence. Less hallucination, more useful answers.

→ In production for a major energy company · 52 challenges in the Academy

Where to see it ↓ Academia Claude Code Energy sector · Confidential
LLMsRAGMCPCopilot Studio
04

Automation & integrations

We connect your tools and delete the repetitive work. What you do by hand today runs itself tomorrow.

→ Running every week: zero manual editing

n8nPower AutomateAPIsWebhooks
06

Technical consulting

We tell you what to build, what to buy and what not to automate yet. Judgement before code.

→ That judgement comes from 16+ systems shipped

Next step ↓ Talk to us on WhatsApp Proposals under NDA
ArchitectureStrategyAudit
[ Selected work ]

Real systems,
in production

C2K C2K — product page: from the table to the kitchen, no paper tickets ● Live (real mode)
8restaurant modules
QRmenu with no app to install
Real timekitchen display and tables
3menu templates
Under the hood

A complete system so the restaurant sells and cooks better: self-service ordering, waiters and tables, delivery with a live map and sales intelligence. Multi-restaurant SaaS: QR menu with 3 photo-first templates, real-time kitchen display with per-step timers, an SVG floor plan with zones assigned to waiters (point-in-polygon), delivery tracking on a live map, batch inventory with FIFO consumption, a CRM derived from the orders and an AI assistant for the owner. Payment is confirmed by a webhook, never by the customer’s browser.

Own product · SaaS / Restaurants

C2K · Cliente to Kitchen

From the customer to the kitchen without friction: QR menu, orders on screen, tables, delivery and inventory.

ReactFirebaseMapLibreWhatsApp APID3
See it live
IICCA A farmer herding livestock down a rural road — sustainable sheep and goat farming ● Live
Voicerecords with no internet
7studies designed
5PDFs generated
Gateprivate area
Under the hood

It also carries the study design and the executable plan, behind a private area. A multi-page static site with a public showcase and a password-gated area (SHA-256 hash on the client), PDFs regenerated with Playwright + Chrome print, and a research design of seven studies with hypotheses and statistical techniques. Hosted as an extra Hosting site, without touching the agency landing page.

Client · AgTech / Development aid

IICCA · Sheep & goat farming

Sustainable sheep and goat farming for the Canada–Latin America Alliance: the smallholder records the flock by speaking, with no internet.

HTML/CSSFirebasePlaywrightPDF
See it live
APPministrador ● Live
Duescollected automatically, arrears down
Hundredsof residents on one channel
AGMbooks clear and up to date
Multibuildings (multi-tenant)
Under the hood

Collect dues on time and bring arrears down with automatic reminders; answer hundreds of residents, owners and vendors without dropping a single complaint; and walk into the annual meeting with the books clear and everything in plain sight. You negotiate with vendors on data, not hunches — and because it is multi-tenant, you multiply capacity without multiplying headcount. One screen so nothing slips: receivables, finance, contracts, bookings, visitors, documents and communication with hundreds of residents, boards, owners, staff and vendors — plus an AI assistant and modules for EV charging, doors and lighting. Multi-tenant SaaS with a super-admin console that enables and bills modules per building; Firestore is closed to the client (everything goes through an API that validates the token). CI/CD deploys on every push.

Own product · Command centre · PropTech

APPministrador

The command centre for property managers: receivables, residents and clear books for the annual meeting — across several buildings at once.

Vue 3TypeScriptFirebaseCloud FunctionsPWA
See it live
Chronos Chronos — hundreds of hours back every month: payroll run with paystubs sent ● Live
529automated tests
ADPone-click payroll export
EN/ESbilingual out of the box
×1.5automatic FLSA overtime
Under the hood

Multi-tenant SaaS where the client never touches the database: everything goes through an API that validates role and organisation. Geofenced clock-in, configurable work week, FLSA ×1.5 overtime, PTO requests and approvals, payroll runs by day/week/biweekly/month with an anti-double-pay lock and a paystub per run, ADP and QuickBooks export, a subscribable calendar (iCal) and a platform console to activate or suspend clients. 529 automated tests, all green.

Collaboration · SaaS / Time & payroll

Chronos

The all-in-one platform for hourly businesses in the US: the team clocks in and out, the system builds shifts, PTO and overtime according to labour law, and the admin runs payroll for any period — with commissions, deductions and the file ready for ADP. In English and Spanish.

Vue 3TypeScriptFirebasePWAi18n
See it live
Predio Radar Predio Radar — map of neglected parcels with a Neglect Score (Gwinnett, GA) ● Live
108kparcels mapped
0–100Neglect Score
5public signals
CSVexport from the dashboard
Under the hood

An off-market radar for neglected property: it finds and ranks parcels showing signs of abandonment. A geospatial join of county parcels with public records (tax delinquency, liens, absentee or out-of-state owners, vacancy) computes a Neglect Score per parcel and paints it as a choropleth on MapLibre, with filters and CSV export. PostGIS backend.

Own product · Real estate / GIS

Predio Radar

Finds abandoned property in a US county by joining public data, scoring each parcel 0–100 on an interactive map.

MapLibrePythonPostGISOpenStreetMap
See it live
atletaone atletaone — performance showcase for a youth athlete In progress
2layers (showcase + lab)
Videoof real plays
Profilewith the athlete’s metrics
Under the hood

A showcase inspired by elite sports sites, sitting on a data layer: per-match metrics, video of real plays and the athlete’s technical profile, designed to grow with wearable data.

Sport · Performance

atletaone

A bioprofile and performance platform for a youth athlete: a public showcase built on a performance data layer.

WebDatosVideo
See it live
Historia Médica Historia Médica — landing page ● Signed app
E2Esync
multidevice
APKsigned and served
Under the hood

Offline-first sync with conflict resolution and blob storage (images and lab results). A signed APK served from our own hosting, working around distribution limits. Each person only ever sees their own history.

Health · Own product

Historia Médica

A portable medical history: lab results, diagnoses and images in one place, with end-to-end sync and a signed APK distributed from our own hosting.

ViteCapacitorFirebase
See it live
Solar + BESS Sol y Red — the estimator’s verdict: investment, first-year savings and 25-year gain ● Live
NASAreal irradiance for the site
PaybackIRR · NPV · LCOE
Off-griddetects if the grid is near
Mapspaste the link and go
Under the hood

A solar-plus-storage self-generation estimator that uses the tariff of the utility that actually serves you. It takes an address (or a Google Maps link resolved server-side), pulls the site’s real irradiance from NASA POWER, measures the distance to the power grid with OpenStreetMap to detect off-grid areas, sizes the PV + battery system with a minimum kit, and runs a 25-year cash flow with Colombia’s Law 1715 tax shield and the right utility per region.

Own product · Energy / Solar

Solar + BESS estimator

Type your address and it tells you whether panels are worth it: investment, savings and payback, on real irradiance.

ReactViteNASA POWERFirebase
See it live
Autopost A published LinkedIn post Operating
1 clickfrom draft to post
Humanin the loop
Verbatimfull-control mode
Under the hood

LLM generation, a human approval queue and publishing through the LinkedIn API with correct format escaping (LTF) so the post doesn’t break when it renders. Versioned drafts.

Marketing · AI pipeline

Content autopost

A pipeline that drafts with AI, goes through human approval and publishes to LinkedIn. The content ships itself; you only approve.

Claude APILinkedIn APIFirebase
[ Toolbox ]

What’s in
the workshop

Tools that have already been through production — not a wish list. Each family links to the project where you can see it running.

01

Product & interface

What the user touches: screens that load fast, make sense without a manual and behave the same on a low-end phone.

ReactVue 3TypeScriptViteHTML/CSS PWACapacitor · AndroidGSAPCanvas 2DSVG D3MapLibreLeafletEleventy Design tokensFigma / FigJamAA accessibilityi18n ES/EN

Seen in C2K Monitas Chronos

02

Backend & platform

What holds the product up once there are people inside: accounts, permissions, money and deploys that aren’t scary.

Node.jsCloud Functions v2Firebase AuthFirestore Security rulesApp CheckStorageHosting Multi-tenantPostgreSQL / PostGISAPIs RESTWebhooks Payments (Wompi)Transactional emailNetlifyDecap CMS CI/CD (GitHub Actions)Cloud SchedulerDNS & domainsSPF · DKIM · DMARC

Seen in APPministrador Chronos Economía Santander

03

AI, agents & voice

Machines that read, answer and speak — with the source in plain sight and a human approving before anything ships.

Claude APIClaude CodeAgent SDKMCPRAG Prompt engineeringLLM evaluationCopilot Studio Microsoft 365 AgentsFabric Data AgentWhisper (speech→text) F5-TTSPiperQR reading (zxing)Human approval

Seen in Academia 2daVoz Autopost

04

Data, models & territory

From raw data to a decision: long time series, maps, optimisation and forecasts that carry their own uncertainty.

PythonpandasLightGBMMarkov chains Conformal predictionMSTLClusteringWeibull oemof.solphHiGHS · MILPSQLETL Power BIMicrosoft FabricArcGISOpenStreetMap ParquetScrapingAPIs XM · SIMEM · NASA POWER · datos.gov.co

Seen in Predio Radar Predictor Energy dispatch

05

Automation & operations

What runs on its own when nobody is watching: scheduled jobs, browsers that work overnight and one-line deploys.

n8nPower Automatepac CLIPlaywright FFmpegCron / Schedulergcloud CLIFirebase CLI GitHub CLIPowerShellBashADBSecrets management

Seen in Market brief The Fleet Autopost

06

Growth & measurement

Getting found — in Google and inside the AI assistants — and then knowing who arrived, from where and what they did.

GA4Search ConsoleBing WebmasterTechnical SEO AEO / AI citationJSON-LDOpen GraphIndexNow llms.txtsitemap · robotsLinkedIn API WhatsApp Business APIMeta AdsSales copy

Seen in Writing Autopost Inteleca

07

Quality, security & judgement

What keeps it from falling over on the first Monday: tests, role-based permissions, data that doesn’t leak, and the decision not to build when building isn’t worth it.

VitestIntegration testsMSWSecurity audit Role-based accessEnd-to-end encryptionAnonymisation Code reviewGit / GitHubDocumentation ArchitectureProduct management

Seen in Chronos Historia Médica APPministrador

[ How the workshop was built ]

One capability
after another

No tool arrived because it was fashionable: each one showed up because a project demanded it. This is the real order, what each one is for and where it was first used.

New capabilities per period
  1. Baseline · up to 2025

    Learning to read data

  2. Python & public-data APIs

    Pulling power-market figures straight from the official source, instead of waiting for a file somebody exports by hand.

    First used in: the price forecaster

  3. Analysis and decision dashboards

    Turning decades of time series into something a committee can argue about without opening a spreadsheet.

    First used in: the demand model

  4. April 2026

    The first product of our own

  5. React + Vite

    The interface: what the user sees and touches, and which has to answer in the first second or it’s lost.

    First used in: La Gran Poya

  6. Firebase: Auth, Firestore, Hosting

    A backend with no servers to administer: accounts, data that updates live and publishing in minutes.

    First used in: La Gran Poya

  7. Capacitor & Google Play

    The same web codebase ships signed as an Android app and reaches the store — without writing the app twice.

    First used in: La Gran Poya

  8. A showcase of our own (static site)

    Having somewhere to show the work before the first client exists. Trust is also built by publishing.

    First used in: the personal portfolio

  9. May 2026

    More products, more rigour

  10. PWA, QR and maps

    It works with nothing to install, and two strangers can find each other to swap what the other is missing.

    First used in: Monitas 2026

  11. Offline sync and sensitive data

    The medical history lives in your pocket even with no signal — and only its owner can see it.

    First used in: Historia Médica

  12. Predictive models with uncertainty

    Forecasting the price and saying how confident we are: a calibrated interval (LightGBM + Markov + conformal prediction), not a promise.

    First used in: the price forecaster

  13. Vue 3 + TypeScript

    Types that catch the bug in the editor, before it reaches a real person’s paycheck.

    First used in: Chronos

  14. Automated testing

    529 tests that run on every change: today’s fix cannot break tomorrow’s payroll.

    First used in: Chronos

  15. Bilingual out of the box (i18n)

    A product born speaking English and Spanish, with language as data instead of a duplicated copy.

    First used in: Chronos

  16. Quantitative market signals

    Explicit rules instead of gut feel to watch a volatile portfolio: opening range, VWAP mean reversion and relative strength.

    First used in: intraday signals

  17. June 2026

    Platform, territory and voice

  18. GIS: PostGIS + MapLibre

    Joining public records on the map and ranking 108,000 parcels by a signal nobody had looked at together before.

    First used in: Predio Radar

  19. Art direction and brand

    A company that builds for others needs a visual voice of its own — type, colour and motion with judgement, not a template.

    First used in: this site

  20. Scientific APIs (NASA POWER)

    Real irradiance for the exact coordinates of the roof, not a country average: the difference between a quote and a guess.

    First used in: the Solar + BESS estimator

  21. Applied geoscience

    Searching backwards: first the hot water and the fault line, then the parcel sitting on top of them.

    First used in: Onsen · Valle de Tenza

  22. Multi-tenant SaaS

    Many clients on one system, each seeing only their own, with a console that activates and bills per client.

    First used in: APPministrador

  23. Payments, webhooks and real time

    The server confirms the payment — never the customer’s browser — and the kitchen sees the order in the same second.

    First used in: C2K

  24. Voice: cloning and synthesis

    Giving someone their own voice back after surgery — and making sure that voice never leaves the house (Whisper, F5-TTS, Piper).

    First used in: 2daVoz

  25. Teaching what we learned

    A course of our own with recall challenges: if a technique can’t be explained to someone else, it isn’t understood yet.

    First used in: Academia Claude Code

  26. Agent architecture

    Splitting the company’s work across agents with a trade of their own, coordinated by an orchestrator (Agent SDK + n8n).

    First used in: The Fleet

  27. July 2026

    And now, being found

  28. Document copilots

    An assistant that answers from the company’s own documents — not the internet — and cites where each thing came from.

    First used in: energy sector · confidential

  29. Mathematical optimisation (MILP)

    Deciding whether to sell or burn the fuel with energy supply as a hard condition, and reading the shadow price of each carrier.

    First used in: the energy hub dispatch model

  30. Editorial CMS (Eleventy + Decap)

    The newsroom publishes on its own, from the browser, without calling the developer to change a headline.

    First used in: Economía Santander

  31. Measurement (GA4)

    Knowing where people come from, what they touch and where they leave — before having an opinion on what to change.

    First used in: this site

  32. SEO and AEO

    Getting indexed by Google and cited by AI assistants: two different jobs, with different rules.

    First used in: the essays

  33. Brand infrastructure

    Our own domain, corporate email that doesn’t bounce and sending authentication in order (SPF, DKIM, DMARC).

    First used in: digiautom.com

[ Writing ]

What we research,
published

[ Academy ] ● New

We don’t just
build them:
we teach you.

Academia Claude Code — an interactive course, taught in Spanish, to learn artificial intelligence and how to really use Claude Code. You learn by doing, with recall challenges and a community.

6modules
27topics
52challenges
5 hof course
  • 00 AI fundamentals
  • 01 Meet Claude Code
  • 02 Context is gold
  • 03 Control and trust
  • 04 Make it yours
  • 05 Advanced automation

[ The record, in numbers ]

16+

systems and prototypes shipped

6

apps in production and in the stores

10

industries: energy, health, insurance, restaurants, real estate, sport…

24/7

machines running with nobody watching

[ How we work ]

From the idea
to autopilot

01

Understand

We talk about your bottleneck, not about technology. The problem first, the tool after.

02

Design

We define the simplest solution that works. Sometimes the best answer is not to automate yet — and we say so.

03

Build

Real software: versioned, tested, documented. No demos that don’t scale and no scripts that only run on your machine.

04

Operate

We leave it running, monitored and in your hands. We teach you to operate it so you don’t depend on anyone.

[ Frequently asked ]

What people
ask us

What exactly does Automatizaciones Digitales do?

We build custom software, predictive models and AI agents out of Bogotá, Colombia: more than 16 systems shipped, 6 apps in production and in the stores, across 10 industries (energy, health, insurance, restaurants, real estate, sport…). We don’t hand over prototypes: every system is left running on its own, versioned and documented.

How much does it cost to automate a process?

It depends on scope, which is why we don’t sell generic packages: a short consulting engagement first defines the simplest solution that works and its fixed budget, before any code is written. Sometimes the honest answer is that automating isn’t worth it yet — and we say that too.

Do you work with small companies or only large ones?

Both. We have built for restaurants, residential buildings, insurance agencies and regional media, and also for a major energy company. The standard is the same at every size: well-modelled data, code that can be maintained and machines that run on their own.

What is an AI agent and what is it good for?

It’s an assistant that reads your data — documents, databases, email — and answers or acts with evidence: it handles customers, builds reports or watches indicators. We run agents in production for a major energy company, and in Academia Claude Code we teach how to build them across 52 hands-on challenges.

Can you forecast prices or demand with AI?

Yes. Our energy price models are trained on 26 years of Colombian power-market data and deliver quantified uncertainty — calibrated intervals, not promises. The same approach applies to demand, risk and inventory.

How do we start?

Write to us on WhatsApp or at hablemos@digiautom.com and tell us where your bottleneck is — the problem, not the technology. We answer within 24 hours.

[ Got something that should be running on its own? ]

Let’s talk.

hablemos@digiautom.com

We answer within 24 hours · Bogotá D.C., Colombia

Message us