Research
Physics through a programmer’s eyes.
We describe the physical world with objects, and only what measured results confirm gets in.
oop-physics-languagephysics.ndot.io01 / 05 · Small test first
Proof of ConceptPoC
in plain words: a feasibility test. A small, fast trial of an idea before you spend serious money on it.
First a small test,then the big project.
Before a large development project starts, it pays to test its most critical points in a small, fast mini PoC project. That way you learn whether the idea works in practice before the big investment. It lowers the risk and helps you avoid the costs of a failed project.
Only 1/3
At Microsoft, only about one-third of the ideas tested delivered the expected improvement; one-third made no difference, and one-third made things worse. In other words, two out of three ideas don’t work as planned – better to learn that in a small test than in a big project.
small test → clear decision → big project
02 / 05 · With real data
We run the test onyour real data.
A demo or a slide deck always looks good, because it is built on ideal data. That is why we build the mini PoC on your real data, alongside your existing systems. Practical obstacles surface during the test – missing or faulty data, difficult connections to existing systems, high running costs – and not after rollout, when fixing them costs far more.
When Target expanded into Canada, the company’s own internal review found that only about 30% of the product data entered into its system was accurate (versus 98–99% in the US). In other words, the bad data only surfaced after launch: the company pulled out of Canada, closed all 133 stores and booked a pre-tax loss of about $5.4 billion.
data quality · system connections · running costs
03 / 05 · Cost risk
One in six large IT projectscosts three times as much.
Most IT projects finish roughly within budget. But one in six costs many times the plan – enough to swallow an entire development budget. In the mini PoC we test the riskiest parts first, so surprises show up while little money is at stake.
A study of 1,471 IT projects found that roughly one in six overran its budget by 200% on average, and its schedule by almost 70%. In other words, these projects cost three times what was planned.
small stakes first · big stakes later
04 / 05 · Clear decision
Every test ends witha clear decision.
Before the mini PoC starts, we agree with you on what it has to prove, what counts as success and who decides on the next step. So the test ends not with an impression but with a clear yes or no. A yes is a solid basis for the big project; a no lets you stop it on firm grounds.
Test first, contract later
According to BMW, it has worked this way since 2015: it first orders a paid test from a new supplier, and signs a long-term contract only if the test succeeds. In other words, every big commitment is preceded by a smaller, verifiable step.
success and stop criteria: set on day one
05 / 05 · A “no” is a result
If the answer is no,that saves you money too.
If the mini PoC shows that an idea doesn’t work, you have only lost the cost of the test – not of a full rollout. We document in writing what didn’t work and why, so the decision holds up in front of the board and the learning stays with your company.
McDonald’s tested AI-powered drive-thru ordering with IBM in more than 100 US restaurants, then ended the test in July 2024 without a wider rollout. In other words, a limited test protected it from the risk of a large rollout.
status: stopped – reasons in writing
Don’t believe it.Bring it.
We build it on your data, judge it against criteria set up front, and you get a straight answer: yes, or a no in writing.
Bring us an idea →Team
Fresh ideas,proven experience.
Projects
What worked.And what didn’t.
We tested these ideas in our projects so far. Some worked, some didn’t – we learned from both.
Live
An internal AI for an entire company.
An international consulting firm uses it in its everyday work; a separate connecting layer links it to the company’s existing systems.
Jupiter 16 · Corporate Bridge · ITadviseLive
Search by meaning, not by keyword.
We made an existing database searchable so that it also finds what users phrase differently.
StageLyncLive
Bank statement reconciliation, automated.
Matching bank transactions to accounting entries is done by the system instead of by hand.
Bluequest · ITadviseConcept
The AI does the job, but never sees the data.
The model only receives the instruction; the company’s own system carries out the operation, so sensitive data never leaves the company.
IntraAiPrototype
An AI that remembers.
It has a long-term memory: what it has learned once, it still knows in the next conversation.
ai_homePrototype
It keeps thinking in the background.
While one AI model works on the task, a second one comes up with ideas and reviews the decisions.
ai_homePrototype
It can improve its own code – but only in a closed space.
The new version is built separately; the proven one keeps running until the new one has proved itself.
ai_homeExperiment
A trading strategy tested on historical data first.
With AI, we test it across many market situations before any real money is at stake.
nDotExperiment
No good model without good data.
We built our own data collection system that records every price move and order, so strategies can be tested on real market data.
nDot_crypto_dbExperiment
The machine tunes the strategy itself.
Using reinforcement learning, it tries out settings and learns from its own results which ones work best.
RlTaTuningExperiment
It looks for market overreactions.
A sudden price move is often followed by a pullback; the strategy looks for these short, recurring situations.
nDot_AcReAExperiment
Same product, two different prices.
The system looks for temporary price gaps between markets: it buys where it is cheaper and sells where it is more expensive, at the same moment.
ndotn_crypto_arbitrageExperiment
Autonomous operation.
The system watches the market day and night and reacts to a sudden move within seconds, without human intervention.
BinanceFastCorrectionExperiment
From research straight into live execution.
A dedicated server carries out the decisions of the strategies developed in the research tool, quickly and reliably.
ndot_trade_serverLive
Job interviews conducted by AI.
We built it into an existing professional community site – no new system needed.
StageLyncAnd many more…
Method
Four steps fromthe idea to live operations.
01
Handing over the idea
With us, handing over an idea is a deliberate process that protects your interests. In the first meetings we get to know the idea and also set the framework down legally. We have an established practice for handling exclusive ideas: depending on your needs, the exclusivity can cover a period of time, the project, or even an entire industry. This creates the safe framework in which the PoC can work.
- Discovery
- Exclusive handling
- Time, project, industry
02
What has to be proven?
Together we pick the most critical points of the idea and agree up front on what counts as success, when we stop, and who decides on the next step.
- The critical points
- Success and stop criteria
- Business owner
03
It works in the lab
We build the mini PoC on your real data, alongside your existing systems. We hand over the results: the code, the documentation and the written learnings.
- Prototype on real data
- Code and documentation
- Written learnings
04
It works in your operations
If the test succeeds, we take it live as a pilot and then hand it over to your team. If it doesn’t, we put in writing why not – so the decision is well founded.
- Pilot
- Handover
- Go-live – or a documented “no”
Live
What does it take for a PoC to work in live operations?
- A real business problem, not technology for its own sake.
- Developers who understand the business goal and its context.
- Success and stop criteria set up front.
- Real, good-quality data.
- A plan for going live from day one: systems, operations, cost.
- People and processes, not just technology.
- A deliberately chosen idea has to be seen through – to failure or to success.
Live.
Live
What does it take for a PoC to work in live operations?
- A real business problem, not technology for its own sake.
- Developers who understand the business goal and its context.
- Success and stop criteria set up front.
- Real, good-quality data.
- A plan for going live from day one: systems, operations, cost.
- People and processes, not just technology.
- A deliberately chosen idea has to be seen through – to failure or to success.
“I’ve learned that a project stands a real chance when it is deliberately focused and gets to the point quickly.”
A company has to keep responding to the changes around it, which means rethinking how it works, again and again. Along the way, we often overestimate or underestimate the market, the future, or our own internal needs. When that happens, we tend to fall back on our own assumptions and set project directions without proper analysis. What I’ve learned is that a project stands a real chance when it is deliberately focused and gets to the point quickly. This is where a PoC helps: it makes clear what we focus on, what we want to achieve, and where the project’s boundaries lie. That is why money spent on a PoC can pay for itself several times over in the larger project, in time and cost saved.
Iván Hőnis, Founder, nDot.io
Got a good project?Let’s work on it together.
info@ndot.io · New Jersey · Budapest · Biel