— LuchoLabs.dev · The Lab · v1.0

lucholabs.dev/lab

I build AI-powered systems and automation infrastructure.
let's talk about:
Bogotá · Remote · C1 English (IELTS Academic 7.0)
Manifesto — what runs through every project

How I Build.

01.Systems Thinking
Real automation isn't writing scripts. It's finding the seam where a workflow leaks — friction between APIs, hidden manual steps, missing observability — and rebuilding it. The open-source live-captioning system in my thesis didn't optimize a model; it closed a tooling gap that locked hearing-impaired students out of lectures. The same lens runs through every project here.
02.AI as Infrastructure
LLMs aren't the product; they're a runtime — and a runtime is only as good as the context you feed it. The real infrastructure is everything around the model: retrieval, grounding, shared memory, and the trust boundary that decides which inputs an agent is allowed to act on. My thesis didn't fine-tune a speech model; it wrapped a generic one in a custom dictionary engine so it actually worked in a Colombian classroom. SUPERFARM is the same bet — a persistent shared-context model the agents reason over. Context is the engineering. The model is the easy part.
03.Cross-Domain Vision
Accessibility tooling adopted by two Colombian institutions. A WhatsApp-first utility-disruption layer for Bogotá. Local-first finance over Colombian bank exports. Phishing protection designed for elderly users. The shape isn't a vertical; it's a refusal to over-specialize when the plumbing is missing everywhere.
04.Builder-Lab Identity
/lab is what it says. One experiment is in active development (Boveda). Others are still design or research (Anti-Phishing Shield, SUPERFARM). Some are archived because the math didn't work (HayCorte, Klipper Copilot — postmortems below). The credibility isn't a hit rate; it's the willingness to write the postmortem.
Thesis — the project that taught me cross-domain plumbing

Captioning Lectures.

Open-Source Live Captioning System for Hearing-Impaired Students
Thesis work 2022–2023 · B.Sc. Systems Engineering · Universidad ECCI · Graduated 2024
Meritorious · In Use
Adopted bySENA Regional Amazonas · Universidad ECCI
StackNode.js · Mozilla DeepSpeech · NGINX · Custom Dictionary Engine
PublicationECCI institutional repository ↗
Built across 2022–2023 after a hearing-impaired classmate had no way to follow lectures in real time. The system generates live subtitles from the teacher's voice directly to the student's screen and stores a full transcript for later review. Completed before the 2024 graduation and awarded a Meritorious distinction. Adopted by two Colombian academic institutions. Zero licensing cost. Deployable by any institution on open source software alone.
— the project that taught me what cross-domain plumbing really means.
Stack — primary runtime + tooling

The Tools.

Core /

  • Python
  • FastAPI
  • n8n
  • REST APIs
  • Linux

Tooling /

  • PostgreSQL
  • SQLAlchemy
  • Webhooks
  • Postman
  • Make
Credentials —AI Automation Engineering · Coderhouse · 2025
Projects — three bets on AI as infrastructure

Three Ecosystems.

BVD · 06

Boveda

Fintech · Local-first OS
Personal finance dashboard for Colombian bank accounts — local-first, programmable, yours.
Dashboard → financial brain
PHS · 07

Anti-Phishing Shield

Cybersecurity · Accessibility
Planned phishing protection for elderly users — plain-language warnings, family-installable. Design stage.
Single check → ISP-level rollout
SFM · 08

SUPERFARM

Agri-AI · Orchestration
Open agricultural intelligence infrastructure — modular runtime for specialized agricultural agents.
Harness → operating layer for agri
BVD · 06 — Project

Boveda.

In Progress
Personal finance dashboard for Colombian bank accounts.
"Banks own your transactions. They shouldn't own how you understand them."
A local-first fintech experiment exploring structured financial data, rule-based automation, and programmable workflows built on top of Colombian bank statement exports.
  • structured imports from Colombian bank statement exports
  • rule-based categorization (no ML required for the boring 90%)
  • programmable workflows on top of your real ledger
  • local-first storage — your data stays on your machine
Long-term →local imports → rule layer → query DSL → fully programmable personal finance
— stack: Python · FastAPI · SQLite · SQLAlchemy · React
PHS · 07 — Project

Anti-Phishing Shield.

Concept
Planned open-source phishing protection for elderly users.
"Security tools built for engineers fail the people who need them most."
A planned open-source tool — at the design stage, not yet in development — to help elderly users identify and avoid email phishing attacks. The design: analyze suspicious emails and produce plain-language warnings non-technical users can act on.
  • plain-language risk explanations (no jargon, no scores)
  • UX tuned for third-age users — large type, slow defaults
  • open source so families can run it themselves
  • email-pattern analysis without sending content to third parties
Long-term →single-email checks → email-account integration → family-installable forwarding inbox → ISP-level rollout
— planned stack: Python · Open Source · Email Analysis · Accessibility
SFM · 08 — Project

Superfarm.

Research
Open agricultural intelligence infrastructure.
"What if agricultural intelligence itself became modular, open, inspectable, and accessible?"
An open-source runtime and orchestration harness for agricultural AI agents — not a chatbot, not a farm app, not another AI wrapper. A modular execution environment where specialized agricultural agents collaborate on shared farm memory, sensors, workflows, and historical context. A farm is a living operational ecosystem — climate, soil, water, animals, crops, finance, logistics, and decisions continuously affecting each other. The intelligence layer should be one too.
  • agent runtime — specialized agricultural agents (SoilScientist, WeatherAgent, HydroponicsAdvisor, ChickenFarmer, CompostEngineer, FinancialAdvisor, IoTDesigner, ScaleUpStrategist) collaborate on the same farm
  • shared farm context — geography, climate, resources, infrastructure, finance, telemetry, and operational history as one persistent model
  • brain router — Claude, OpenAI, Gemini, Ollama, local models; pluggable per task
  • tool layer — weather APIs, GIS, IoT sensors, telemetry, vector DBs, financial workflows
  • open source by design — local deployment, inspectability, technological sovereignty
Long-term →runtime foundation → specialized agents → sensor + telemetry integration → ecosystem expansion → open agricultural intelligence infrastructure
— currently in architectural & research phase. validating runtime, orchestration, shared-context models, and modular agent execution.
"Linux became OS infrastructure. Kubernetes became orchestration infrastructure. ROS became robotics infrastructure. SUPERFARM explores the same shape for agriculture."
The Lab — recent commits + what's next

Now & Next.

Now /

Boveda · Anti-Phishing Shield · Automation Workflows

Next /

[writing] Cornerstone build log: why local-first finance — building Boveda after Colombian banks broke me.

[talk · past]AI Tinkerers Bogotá — Cámara de Comercio, May 2026 — Lightning talk on the organizational context layer
[talk · past]AI & Cybersecurity Forum — Bogotá, May 2026 — Panelist: "Cybersecurity in the Golden Era of AI"
[talk · past]Cybersecurity Forum — Bogotá, 2025 — Panelist on network security and infrastructure
Archive — postmortems from experiments that died

The Graveyard.

total 2 drwx------ 1 lual staff haycorte/ drwx------ 1 lual staff klipper-copilot/
"Most experiments die. The ones that live earned it."
Killed haycorte urban disruption intelligence — bogotá

What it was

WhatsApp-first infrastructure intelligence platform for utility outages (water / electricity / gas) in Bogotá. Bridged the gap between official utility info and how residents already share information in WhatsApp groups.

Why it's strong

Real communication problem; the gap is real.

Why it died

  • Infrastructure dependence. WhatsApp automation, utility scraping, geolocation, cloud APIs — none of it under your control. Continuous reliability required from day one because users only show up during stressful events; one scraper failure or blocked WA flow burns trust instantly.
  • Business-structure mismatch. Behaves like a public-utility layer; monetizes like SaaS. Individuals won't pay for outage info. The strongest path (building admins / property mgmt) needs manual sales, onboarding, local ops.
  • Episodic engagement. Users only care during outages. Retention depends on scheduled maintenance alerts and community participation, which delays self-sustainability.
  • Capital vs. operational burden. Infrastructure-heavy with insufficient early revenue to justify the load. Without capital or a dedicated ops layer, it operates at a loss for too long.
— strategically interesting as a future utility-data layer. for now: archived.
— · —
Killed klipper-copilot ai companion for orcaslicer + klipper

What it was

An AI-assisted companion layer for OrcaSlicer and Klipper — analyzing slicer profiles, validating printer constraints, and recommending safer print settings for high-performance 3D printers.

Why it's strong

Real systems-engineering pain; tuning these printers is genuinely hard.

Why it died

  • Infra cost > realistic revenue. Reliable optimization needs large datasets of printer configs and failures, telemetry ingestion, hardware-specific tuning models, and continuous testing across many printer variants. Support burden disproportionate to likely revenue.
  • Audience economics. Target users (advanced hobbyists) are technically capable, price-sensitive, and resistant to subscriptions — hard acquisition and retention without a large community presence or distribution channel.
  • Trust-fragile category. Inconsistent recommendations or one failed print suggestion burns trust fast. High API and hosting costs while debugging edge cases make the early stage punishing.
  • Strategic fit. Better classed as a systems-engineering experiment than a product right now; infrastructure and validation requirements exceed probable short-term return. A loss-making launch is strategically unrealistic at this stage.
— technically interesting; product-market gravity insufficient. archived.
Proof — credentials and adoption

Trust at a glance.

role→Freelance Automation Engineer · Self-employed · Apr 2024–Present
role→QA Tester · Funktronic Labs (Remote) · 2024
cert→AI Automation Engineering · Coderhouse · 2025
thesis→Meritorious thesis project · built 2022–2023 · graduated 2024 — adopted by SENA + ECCI
full→View full CV ↗
Speaking — talks and panels

Speaking proof.

2026→AI Tinkerers Bogotá · Cámara de Comercio · May 2026 · Lightning talk 2026→AI & Cybersecurity Forum · Bogotá · May 2026 · Panelist
2025→Cybersecurity Forum · Bogotá · 2025 · Panelist
Connect — channels

Let's Talk.

git→github.com/lualducor in→linkedin.com/in/luis-alberto-duarte-cortes-97748171 mail→lualducor@gmail.com write→lucholabs.dev/blog
— Source · github.com/lualducor/THELAB