00 / MMXXVI · Intro
Operator working where finance, automation & security meet_
- Who
- Luis De Freitas
- What
- Finance, Software, Marketing
- Where
- Jersey, Channel Islands
- How
- Being Passioned
- When
- Available
01 / Recently
Personal Projects
-
What it does
- Opens with four doors: I need help, I want to help, I can carry things, I run a centre.
- Lists every relief centre with what it needs, what it holds and what is already covered, kept current by the centres themselves.
- Signs a centre in with a token and a PIN, so it can update its own needs without an account or a password reset.
- Suggests transfers, moving a surplus at one centre to another that is short, with a route and a contact.
- Follows any donation publicly by its invoice token, without ever revealing a donor name or a phone number.
- Finds missing family by name and puts the nearest centres on a map.
- Installs on a phone and keeps its shell working on a weak connection.
Problem it solves
In the days after a disaster nobody knows who needs blankets, who has too many, or which van is going which way. This puts all of it in one live directory that the centres maintain themselves. No browser writes to the database directly. Every change goes through a single server function behind a rate limit, and identity documents sit in a private bucket. Anyone can trace a donation from the invoice to the centre that received it, and nobody can read a donor's phone number while doing it.
-
What it does
- Replaces a paper coffee card: ten stamps, one free cup.
- Staff raise a single-use QR code that the customer scans with the phone camera.
- Every token is minted on the server, expires after five minutes and burns on first use.
- Gives the shop a live count of cups given, customers served and the week's trend.
- Exports weekly and monthly redemption reports as PDF.
- Separates customer, staff and admin rights at the database level.
Problem it solves
A paper stamp card is trivial to forge and impossible to measure. The obvious digital version, a button the customer taps, is worse. Here a stamp exists only if the server issued a token for it, that token dies after five minutes and cannot be spent twice, so a customer cannot mint their own free coffee. In return the shop gets what the paper card never gave it: a running count of what it is giving away and what that costs.
-
What it does
- Meets a guest at the door and routes them: eat in, take away, delivery, book a table, or just read the menu, then remembers the choice.
- Reads the QR code on a table and drops that guest straight into the eat-in flow.
- Runs the account layer: registration, sessions, password resets and permissions tied to a role.
- Asks anyone with a management role for a rotating code from an authenticator app.
- Writes every sensitive action to an audit log.
- Models the whole restaurant in the database: tables, bookings, menu, modifiers, allergens, stock batches, orders, kitchen tickets, loyalty, shifts and rotas.
Problem it solves
A restaurant runs on a pile of tools that do not talk: one for bookings, one for the till, one for rotas, and none of them agree at the end of the month. This is the foundation of a single system meant to hold all of it. The identity and permission layer is built and hardened, and anyone who can move money or change a rota needs a second factor to sign in. The data model for the rest of the floor is designed. Everything a guest actually touches is still to be built.
-
What it does
- Derives a minimum day rate from overhead, cost of living, target profit, equipment amortisation and combined tax, then spreads it across the days actually billable in a year.
- Prices each job off a rate card of crew roles, gear, mileage and per diem, then applies an uplift for usage rights: organic, paid social, or broadcast buyout.
- Presents every quote as two figures, strategic value against launch investment, in an Option A and an Option B.
- Flags any quote whose effective day rate lands under the calculated floor.
- Prints an A4 PDF proposal from the live quote and keeps clients, quotes and line items in SQLite.
- Holds all money as whole pence and applies GST once, at the total.
Problem it solves
Creative work is usually priced by instinct, so jobs get sold below the cost of delivering them and nobody notices until the year is closed. The engine separates the two questions a quote has to answer: what the work is worth to the client, and what it costs to deliver. The cost side is computed from the studio's own overheads rather than guessed, so a job that would lose money says so before it is sent.
-
What it does
- A nightly memory engine distils every conversation and file into a searchable knowledge graph.
- A research agent gathers credentialed sources, flags uncertainty and refuses to fabricate data.
- Agents for sourcing, negotiation and price hunting run on a schedule without me.
- Every agent stops at a human checkpoint before anything leaves the system.
Problem it solves
Knowledge earned once was being lost, and the same research repeated month after month. The fleet turns work I used to redo into something that runs on a schedule and leaves a record. Nothing reaches the outside world until I have approved it.
-
What it does
- Continuously audits both what was done in a session and the live state of the host.
- Flags plaintext secrets, over-permissive containers and newly opened ports.
- Quarantines every third-party component and reviews it in isolation before it can run.
- Fails closed: on any error or doubt, nothing is installed.
Problem it solves
Untrusted code and drifting configuration are how systems quietly become insecure. Here the review happens before execution rather than after an incident, and the gate refuses whenever it is unsure, so a mistake costs a rejected install instead of a breach.
-
What it does
- Runs 30+ containerised services on a hardened server, each behind automatic TLS.
- Exposes nothing to the public internet: access is over a private mesh network only.
- Enforces least-privilege credentials, terminal audit trails and immutable logs.
- Hosts local AI models so sensitive data never leaves the machine.
Problem it solves
Most self-hosted stacks are one misconfigured port away from exposure. Making the private network the only door, instead of layering a firewall rule over a public one, removes an entire class of mistakes. Running the models locally means confidential material is never handed to a third party.
-
What it does
- A web console to see, start, stop and create every agent running on the server.
- Shows live logs, schedules and whatever each agent produced overnight.
- Built with Next.js, served privately over the mesh network.
Problem it solves
Before this, checking on the agents meant reading log files over SSH and guessing at the rest. The console puts what ran, what it produced and what failed in one place. I would not leave anything running overnight without it.
-
What it does
- Splits into two views over one database, one for the crew on site and one for the office.
- Clocks builders in against a project and stamps the entry with GPS coordinates.
- Reads a supplier invoice from a photo and files it as structured line items against the job.
- Imports a task list straight from a client spreadsheet, grouped by section and room.
- Holds signed risk assessments per project, plus daily reports with photos taken on the phone.
- Runs the store for materials and tools: checkouts, transfers, deliveries and waste collection.
- Closes each task with photo evidence the manager approves or sends back for correction.
- Exports hours and invoices to Excel for payroll and cost reporting.
Problem it solves
On a building site the record of what happened lives in paper timesheets, photos on someone's phone and invoices in a van. BuildTrack Pro captures each of those at the moment it occurs and ties it to a project, so hours, spend and safety sign-off land in one auditable place. Row-level security in the database holds a builder to their own work while the office sees the whole site, and new workers join by scanning a QR invitation rather than being added by hand.
02 / Roles
Previously.
03 / Origin
Humble Roots, Ambitious Dreams
It started with a degree in International Commerce and a habit of building. Moving between continents (Venezuela, Spain, then Jersey) added a different kind of fluency.
I taught myself to ship software and, just as importantly, to secure it: hardened infrastructure, least-privilege access, audit trails by default. Then I went after accounting and reporting, the numbers behind a business, so I could connect the systems I build to the decisions they serve.
04 / Education
Trajectory.
05 / Method
Approach.
I start from least privilege and automate whatever repeats. Everything leaves an audit trail, because I would rather hand someone evidence than a promise.
A breach is the maximum form of operational disorder. Security is part of how the work gets done.
06 / Side projects
Tinkering.
Most of the marketing work I do starts as a system before it becomes a campaign. On a VPS I manage, I have built a set of protocols that turn a brief into research, search intent, SEO structure, campaign angles and a working production plan. The process is mine: I wrote it, trained it around the way I work and keep refining it whenever a project exposes a weak assumption.
That same system helps me develop video ideas with a clear job to do. It can move from audience research to a script, visual references, edit notes and cutdowns without losing sight of the action the video is meant to produce. I use it to compare sources, test claims, audit sites and funnels, and find the technical or strategic weak points holding a campaign back. The agents can keep working overnight, but nothing is published, sent or changed without my review.
I keep extending the system into the parts of marketing that usually become fragmented: SEO, audiovisual production, reporting, asset organisation and follow-up. Alongside it, I am building an app to make the organisation and tracking of donations in Venezuela less improvised and easier to coordinate. The tools handle the sorting and repetition. I still decide what is worth saying, what should be made and what the work is for.
07 / Stack
Tools.
08 / Say hello