OSOM Labs · Lausanne · Sion
Ideas come to life
Understand what the web and AI are changing for your business.
Tools to try. Research to follow. Explore what we are building at OSOM Labs.
Explore the experiencesAn answer that brings together information from several pages.
AI proposes a summary. You check the sources.
- 01Choose a question
- 02Understand the mechanism
- 03Try it for yourself
A tool to use
Your site, from the other side
One address. Observations about your page. Practical steps to make it easier to understand.
See how it worksThe steps in detail
Choose a step to read it in detail.
Your page. Enter the address of the page you want to examine. You then confirm the analysis on the audit screen.
An outside view. The audit reads the page’s code (headings, description, links, images) and, when available, its speed as measured by Google, to identify what makes it easier or harder for search engines to understand.
Practical next steps. Read the observations and suggested improvements without having to leave your email address.
An initial look at one specific page of your website, with suggestions to consider.
Prepare my audit
Enter an address to prepare an analysis of your page.
You will confirm the analysis on the next screen. It examines the supplied page, not the whole website. No email required to view the results.
A simulation to adjust
One adjustment. An answer.
Expertise becomes useful when you can interact with it. Adjust the inputs in this calculation demonstration.
See how it worksThe steps in detail
Choose a step to read it in detail.
Your scenario. Choose how long a task takes, how often it comes up over a period of your choice, and the share you think could be automated.
A simple calculation. The calculation multiplies the duration by the number of tasks and the share that could be automated. Minutes are then converted into hours.
A sense of scale. Move the sliders to see what changes. The result illustrates your assumptions; it does not measure savings already achieved.
A rough estimate that does not yet include setup and review time.
Set up my scenario
Move the sliders and see what changes.
Potentially automatable
2 hof time in your scenario
Minutes × tasks × share ÷ 60. An illustration based on your assumptions, before setup and review time.
A method to explore
Found, and described correctly?
Our measurement project: putting customer questions to an AI assistant, reviewing its answers and keeping its sources.
See how it worksThe steps in detail
Choose a step to read it in detail.
A customer question. We start with business questions: finding a provider, comparing options or checking what a business does.
The answer and its sources. Each answer is kept exactly as given, with its sources, its date and the interface queried.
Facts to check. For each answer, a person notes whether the business is named, whether it is recommended, whether what is said is accurate, and which pages served as sources.
Here, you explore our research method; no results are measured live.
Explore the questions
Choose a question to explore our method.
Explore the protocol · illustrative questions, no live results
« Who can help me make my website easier to understand? »
Does the business appear without being named in the question?
See the three steps of the protocol
- Ask
A defined set of business questions, repeated under a recorded protocol.
- Keep the evidence
The original answer, its sources, the date and the interface used.
- Compare
Mentions, citations and facts, with human review of ambiguous cases.
Tried on ourselves, 27 October 2026
Nine questions put to an AI assistant, up to three times each: 25 answers, reviewed one by one. Without our name in the question, OSOM appeared in none of the 17 answers. With our name, the assistant still quoted prices removed from our site on 31 May, and mistook our founder’s brevet fédéral for the higher-level diplôme fédéral. Our English page had translated it loosely; we corrected it the next day. The same questions will be asked again in mid-December, once the search index has caught up.
A principle to discover
Find the right context
An organised memory to retrieve a decision when it matters. The principle, shown with local examples.
See how it worksThe steps in detail
Choose a step to read it in detail.
Decisions. A visual identity, a contact page, a Labs experience: every project carries decisions that need to be retrieved.
An organised memory. The principle is to organise this context so the relevant decision can be found when work resumes.
The right context. Choose a topic in the demonstration to reveal the corresponding decision. The examples are written into this page.
Examples written into the page illustrate the principle.
Open the folders
Open a folder to retrieve an example decision.
Use the yellow dot as the visual signature.
Illustration of context retrieval with local examples. This demonstration does not query the MemPalace engine.
A first-hand account to read
Build. Measure. Question.
A system can look convincing and predict nothing. This project shows how we put its results to the test.
See how it worksThe steps in detail
Choose a step to read it in detail.
An appealing idea. This internal project starts with an idea: build a score to help interpret stock-market signals.
The test of evidence. The score is compared with observations. A convincing presentation is not enough: what the system actually predicts needs to be checked.
An acknowledged limit. The story shows why a verdict that contradicts the original idea must be accepted: the score predicts no better than a coin toss. It shares a method for checking a system, and its limitations. Since then, the system no longer looks for the right moment to buy: it applies a written plan and refuses to advise on incomplete data.
An internal educational case, showing how to put a system to the test.
Examine the approach
Follow the reasoning, then read the story of the experiment.
Build a score that helps interpret stock-market signals.
Compare its predictions with the observations. Accept a negative verdict.
Internal educational project. The full case explains the method and its limitations.
Read the full storyAn app to examine
Between two sets, nothing more.
A custom training app designed for one person, for use at the gym between two sets. An internal project, in real use since summer 2026.
See how it worksThe steps in detail
Choose a step to read it in detail.
At the gym, between two sets. Everything starts with how it is used: a phone, at the gym, between two sets, where logging has to fit into a few seconds. The app installs on the home screen and opens only one exercise at a time, with its anatomical illustration, the target muscle shown in red, and the settings noted for each machine.
One exercise, one tap. The open exercise is already filled in: the planned sets, then the target chosen last time, otherwise what was logged last time. “Done · 3 × 10 · 20 kg” logs all three sets at once. Set-by-set detail, the note and “Next time?” remain optional.
Suggested, never imposed. The rule is double progression: reps first, and a heavier load is only suggested once the weakest set reaches the top of the prescription. The app suggests it; it never applies it on its own. “Next time?” sets the target for the next time this exercise comes up, and that target prefills the log.
A custom app proves itself in use: this one has served at the gym since summer 2026, and its calculation rules are covered by automated tests.
Log an exercise
Log a fictional exercise in one tap, then adjust the sets to see when a heavier load is suggested.
Fictional data · nothing is saved or sent.
1 · One exercise, one tap
Prefilled from the programme. One tap logs all three sets.
2 · Suggested, never imposed
Adjust the reps of the three sets. Prescription: 3 × 10 at 20 kg, 2.5 kg step.
- Set 1
- Set 2
- Set 3
Weakest set 9 / 10
No step up yet: the weakest set has to reach 10 reps. Reps come first.
The weakest set decides, not the best one: 12, 8, 8 says the load is already where it should be.
- Since summer 2026in real use
- 6digits to sign in, no password
- Zurichwhere the database is hosted
- 286automated tests as of 26 October 2026
An idea to build
together?
A precise need can become a clear, measurable and genuinely usable tool.