Expertise, amplified.
Automation doesn’t replace hard-won expertise. It clears the path for talented people to leverage it.
Expertise,
amplified.
Automation doesn’t replace hard-won expertise. It clears the path for talented people to leverage it.
I’m Anthony Calek. Welcome to happycodech.
An independent practice built on cross-domain experience: managing assets, understanding how businesses actually run, and engineering the software that connects them. The result is a measured approach to AI and process automation. Practical, not revolutionary.
I’m Anthony Calek. Welcome to happycodech.
An independent practice built on cross-domain experience: managing assets, understanding how businesses actually run, and engineering the software that connects them. The result is a measured approach to AI and process automation. Practical, not revolutionary.
AMPLIFY, DON'T REPLACE
Automation is a layer over expertise, not a replacement. Pointed at the right problem, it does what people can’t do alone.
SIGNAL IN THE NOISE
Language analysis reads what a spreadsheet can’t: the connections that only surface with the right process.
BUILD, THEN IT BUILDS ITSELF
The right workflow turns fast prototypes into high-quality systems, even software that improves on its own loop.
What I do
Applied AI Engineer
Capabilities deployed.
I build AI systems that reach production and stay trustworthy there: the agent workflows, the data foundation beneath them, the evaluation that keeps them honest, and the judgment for when a model is the wrong tool. Embedded with your team, not advising from a distance.
How the work gets pointed at a problem: learning an unfamiliar domain fast, turning a vague requirement into a system spec, and knowing when the answer is an LLM and when it is plain deterministic code.
Where applied AI has moved past single prompts: supervised agent workflows that check their own output, retrieval grounded in real sources, wired into something that holds up in production.
The unglamorous layer most AI projects skip and then regret: clean ingestion, sources that reconcile, messy documents turned into structured data, a schema that stays trustworthy as it grows.
The discipline that separates a working system from a confident-looking one: measuring whether the output actually predicts what it claims, built so the result reproduces rather than gets taken on faith.
Hover or tap any card to flip it
What I do
Applied AI Engineer
Capabilities deployed.
I build AI systems that reach production and stay trustworthy there: the agent workflows, the data foundation beneath them, the evaluation that keeps them honest, and the judgment for when a model is the wrong tool. Embedded with your team, not advising from a distance.
How the work gets pointed at a problem: learning an unfamiliar domain fast, turning a vague requirement into a system spec, and knowing when the answer is an LLM and when it is plain deterministic code.
Where applied AI has moved past single prompts: supervised agent workflows that check their own output, retrieval grounded in real sources, wired into something that holds up in production.
The unglamorous layer most AI projects skip and then regret: clean ingestion, sources that reconcile, messy documents turned into structured data, a schema that stays trustworthy as it grows.
The discipline that separates a working system from a confident-looking one: measuring whether the output actually predicts what it claims, built so the result reproduces rather than gets taken on faith.
Tap any card to flip it
How I work
Capture what your firm knows. Multiply it. Keep it yours.
For most of business history, the knowledge that made a firm work stayed locked in the people who held it, and walked out the door when they left. That is no longer necessary. I surface what your firm actually knows, multiply it with automation that fits the real work, and make sure you own every piece of it.
The Work Nobody Wrote Down
I find the operating knowledge your firm runs on and never documented: the judgment calls, the workarounds, the way your best people actually do it.
This is the part software ignores and most consultants skip. The judgment that makes you hard to copy sits in a few heads, undocumented and one resignation from gone. I surface it on purpose, because it is both the thing worth automating and the thing worth protecting.
What You Already Own
Before any new spend, I map the capability you already pay for and the tools your team adopted on its own.
Most firms license capability they never switched on, across Microsoft, Google, and their SaaS seats, next to shadow tools nobody governs. The quickest wins usually hide in what you already own; the quietest risks, in what you did not know your team was using.
A Record That Outlives Me
Everything I learn becomes a structured, queryable knowledge base you own, not a slide deck you file and forget.
Discovery leaves an asset, not a report: a living, queryable record of how your operation runs, that your people and your automations both read from. It is what every later build stands on, and it keeps paying off long after I am gone.
Expertise, Multiplied
I turn the knowledge we captured into automation that handles the rote work and hands the judgment back to your people.
Amplification, not replacement. The system takes the repetitive, error-prone, time-sink work and routes the real judgment back to the people who are good at it. Your seniors stop doing junior work, and the more your knowledge feeds the system, the better it gets.
The Simplest Thing That Works
Configure before automating, automate before buying. Each step has to earn the next.
I escalate on purpose, and I stop early. Most problems are solved by switching on what you have or wiring two systems together, not by a new platform. It keeps cost, risk, and the maintenance burden nobody budgets for from compounding.
A Number, Not a Promise
Every build ships against a baseline, so you see exactly what it removed: hours, errors, or days.
I measure against a baseline taken up front, so each build arrives with a before-and-after figure, not a claim: hours saved, errors removed, cycle time cut. The kind of result a CFO signs off on, because it is auditable.
On Swiss Soil, on Your Terms
Model-agnostic, deployed where your compliance team requires. FADP, Berufsgeheimnis, FINMA, and the EU AI Act, designed in.
It deploys where you need it: Azure Switzerland, Swisscom Sovereign Cloud, or EU-only, with no vendor lock. Data residency and the regulatory frame are built in from the first sprint, not retrofitted the week before an audit.
The Knowledge Stays Yours
Every output and reasoning trace lives in your tenant, under your keys. When I leave, there is nothing to hand back.
Nothing runs through my accounts. The captured knowledge, the systems, the audit trail, all of it sits in your tenant under your control. No lock-in, no dependency on me: when the engagement ends, you already hold everything that matters.
Rules You Can Defend
You leave with a written, enforceable policy for where automation belongs and where it does not.
The new way of working needs governance your old policies never anticipated. You leave with a usage policy your firm can follow and a board or regulator can accept: which tools, what data, and where you deliberately keep a human in the loop.
Hover or tap any card for the detail
Three services. One practitioner. Free first conversation, in your office, often same day.
How I work
Capture what your firm knows. Multiply it. Keep it yours.
For most of business history, the knowledge that made a firm work stayed locked in the people who held it, and walked out the door when they left. That is no longer necessary. I surface what your firm actually knows, multiply it with automation that fits the real work, and make sure you own every piece of it.
The Work Nobody Wrote Down
I find the operating knowledge your firm runs on and never documented: the judgment calls, the workarounds, the way your best people actually do it.
This is the part software ignores and most consultants skip. The judgment that makes you hard to copy sits in a few heads, undocumented and one resignation from gone. I surface it on purpose, because it is both the thing worth automating and the thing worth protecting.
What You Already Own
Before any new spend, I map the capability you already pay for and the tools your team adopted on its own.
Most firms license capability they never switched on, across Microsoft, Google, and their SaaS seats, next to shadow tools nobody governs. The quickest wins usually hide in what you already own; the quietest risks, in what you did not know your team was using.
A Record That Outlives Me
Everything I learn becomes a structured, queryable knowledge base you own, not a slide deck you file and forget.
Discovery leaves an asset, not a report: a living, queryable record of how your operation runs, that your people and your automations both read from. It is what every later build stands on, and it keeps paying off long after I am gone.
Expertise, Multiplied
I turn the knowledge we captured into automation that handles the rote work and hands the judgment back to your people.
Amplification, not replacement. The system takes the repetitive, error-prone, time-sink work and routes the real judgment back to the people who are good at it. Your seniors stop doing junior work, and the more your knowledge feeds the system, the better it gets.
The Simplest Thing That Works
Configure before automating, automate before buying. Each step has to earn the next.
I escalate on purpose, and I stop early. Most problems are solved by switching on what you have or wiring two systems together, not by a new platform. It keeps cost, risk, and the maintenance burden nobody budgets for from compounding.
A Number, Not a Promise
Every build ships against a baseline, so you see exactly what it removed: hours, errors, or days.
I measure against a baseline taken up front, so each build arrives with a before-and-after figure, not a claim: hours saved, errors removed, cycle time cut. The kind of result a CFO signs off on, because it is auditable.
On Swiss Soil, on Your Terms
Model-agnostic, deployed where your compliance team requires. FADP, Berufsgeheimnis, FINMA, and the EU AI Act, designed in.
It deploys where you need it: Azure Switzerland, Swisscom Sovereign Cloud, or EU-only, with no vendor lock. Data residency and the regulatory frame are built in from the first sprint, not retrofitted the week before an audit.
The Knowledge Stays Yours
Every output and reasoning trace lives in your tenant, under your keys. When I leave, there is nothing to hand back.
Nothing runs through my accounts. The captured knowledge, the systems, the audit trail, all of it sits in your tenant under your control. No lock-in, no dependency on me: when the engagement ends, you already hold everything that matters.
Rules You Can Defend
You leave with a written, enforceable policy for where automation belongs and where it does not.
The new way of working needs governance your old policies never anticipated. You leave with a usage policy your firm can follow and a board or regulator can accept: which tools, what data, and where you deliberately keep a human in the loop.
Tap any card for the detail
Three services. One practitioner. Free first conversation, in your office, often same day.
How I work · a worked example
I built it the way I would build yours.
Language Analysis is a system I built to read what companies stop saying in their SEC filings. I will follow one real signal the whole way through, Molina Healthcare’s 2020 annual report, from a question on a page to a documented product you could own, using every capability above. If you want, I will walk you through it one phase at a time, and you will see the actual machine work as you go.
Phase 01 · Discovery
Frame the question, not the code
Before any code: when does a company’s filing language quietly drift from its peers, in a way a person should read? I wrote down the question, the data model, and a reason for every decision. To keep it concrete I will follow one filing that did exactly this, Molina Healthcare’s 2020 10-K. It is a monitoring tool, not a trade-signal generator.
0
filings read
0
language flags
0
sectors
Phase 02 · Build
Stand it up, then trace one company
Within a day it was a working, automated tool: one pattern across three sectors, run by one person. A new sector is a config file, not new code. Here is Molina run through it, eight annual reports, each a dot sized by how many language flags fired. 2019 is silent. In 2020 the language erupts to sixty-six flags, the only Molina year flagged against the entire corpus.
Phase 03 · Discovery · the signature
I tried to break it
Seven kinds of flag, four independent tests. Watch six of them fall away.
look-ahead
multiple hyp.
earnings
placebo
Margin of the term-disappearance signal over a random-word stand-in, managed care. Single placebo permutation, one seed, managed care only. I do not claim this across sectors.
Phase 04 · Build
Find the real bug, build the real fix
Molina’s Risk Factors did not arrive clean. The same 10-K section hides in different document structures depending on which filing software built it. I built a reader for each dialect, plus an unknown bucket held for human review, so a section like Molina’s gets pulled by the right rules. Hover or tap a lane to see the receipt every extraction carries.
Phase 05 · Secure
Every result carries a receipt
Here is the actual Molina flag, with the stamp that lets any result trace back by one query. And the honesty runs deeper: my best-looking result had a bug in it. Click the card.
New language that appeared in 2020, absent the year before
The same discipline, turned on my own results:
My best-looking result
65.11% hit / +4.37% return
90-day, before I checked it hard enough.
Click to see what was real →
After I found my own bug
42.20% hit / -5.60% return
I named the wrong number and retired it in the published paper. The honesty is the product.
Phase 06 · Build
How I’d apply this to your firm
The same pipeline that read Molina’s 2020 10-K can read yours. It ends as a documented product a person runs from a runbook. Here is the sequence I would run on your filings:
- Frame your question and data model before any code.
- Stand up a working sector pipeline with alerts.
- Test it to destruction and report only what survives.
- Diagnose extraction failures to root, build for your filers’ structures.
- Stamp provenance, govern the data, ship a clean release you own.
- Document it as a transferable product.
The next layer, a peer-aware semantic search that compares filings by meaning rather than vocabulary, is scoped and feasibility-checked but not yet built. I will say so plainly until it is.
Same method, your firm. The first conversation is free, in your office, often same day.
The live project
Language Analysis, in production
Three sectors live, the methodology documented, the paper published. See the real thing running, not a description of it.
Get in touch
A free first conversation. In your office. Often same day.
Most of my work starts with a short conversation. If you have a small technical problem you don’t know what to do with, or a larger problem you’re not sure how to scope, that conversation is the right next step. There’s no obligation and no pitch. I’ll tell you whether I’m the right fit, and if I’m not, I’ll usually know who is.