English is not my first language. I used AI to help translate and edit parts of the proposal, but the project idea, research plan, architecture, implementation decisions and claims are mine. I take responsibility for the final text and for the work behind it.

Working prototype demo: https://www.youtube.com/watch?v=1JI98fcirps

Project summary

xFilosofem started from a problem I kept noticing while I was studying different things.

I could know something in one situation and fail to use it in another. In Russian, for example, I could recognize a word but not produce it, or understand a grammar rule and still use it incorrectly in a new sentence. I started wondering whether a learning system could keep track of these differences instead of reducing everything to right or wrong.

The idea is to record what the learner actually does while using the system: their answers, choices, mistakes, use of hints, and how these change over time. I want to use this history to look for patterns that the learner may not notice themselves.

I also do not want xFilosofem to stay inside one subject. Russian is the first main domain, but Medicine is planned as a second and very different domain. The long-term goal is to understand which parts of learning are specific to a subject and which patterns, if any, can be useful across different subjects.

The system should not simply decide that it “understands” the learner. Its predictions should be compared with what actually happens later. If the predictions are wrong, the model should change.

What are this project's goals? How will you achieve them?

The next goal is quite simple compared with the long-term vision.

I want xFilosofem to make a prediction before I answer a task, then compare that prediction with the real result. The first version of this will be in Russian, because that is where the system already has the strongest working material and history.

I do not want to assume that a complicated model is automatically better. I want to compare it with much simpler approaches, such as recent performance, repetition history, and time since the last attempt. If the simpler model works just as well, then that is the model I would rather keep.

After the Russian loop is working, I want to introduce a small Medicine domain. Medicine is useful because it creates very different problems: laboratory findings, mechanisms, clinical cases, incomplete information, and situations where the correct conclusion and the reasoning behind it are not always the same thing.

One of the main questions for me is whether anything useful can actually be shared between these domains. I do not expect Russian and Medicine to use exactly the same model. Some parts may stay completely separate. I want the system to test that rather than decide it in advance.

The basic cycle is:

Task → Response → Evaluation → Prediction → Real Outcome → Pattern → Teaching Decision → New Evidence

The system keeps track of what the learner does and makes predictions about what they will be able to do later. Over time, those predictions are compared with what actually happens, and the picture of the learner can become more accurate.

Adaptive teaching is a later step. First I want to know whether the system can make useful predictions at all.

How will this funding be used?

The biggest thing this funding would give me is time.

I am studying medicine while building xFilosofem, and at the moment I work on the project around my university schedule. I can keep doing that, but it makes progress much slower than it could be. Having some financial breathing room would let me spend more consistent time on the project instead of only working on it in whatever time is left.

The funding would also cover the software, AI/API, hosting, and other infrastructure I use while developing and testing the application.

Another part would go toward independent review of the first Medicine material. I am a medical student, but I do not want medical questions or evaluation rules to be accepted simply because I wrote them myself. I want outside review before I make any serious claims based on that part of the system.

There will also be practical costs related to the first real prediction tests and later evaluation.

If the project only reaches the minimum funding level, I would focus on the Russian prediction loop and the first proper evaluation. With more funding, I could spend more time on the Medicine side, independent review, longer prospective testing, and the cross-domain comparison.

For me, the important part is that xFilosofem is reaching a point where it needs more than spare-time development. The next stage is not only adding features. It is where the research part of the project starts becoming real.

Who is on your team? What's your track record on similar projects?

I am a medical student, and my project partner is an electrical and electronics engineer.

Most of the architecture, product direction, and learning-system design in xFilosofem have come from me. We have worked together on coding, testing, and reviewing the application.

This is my first project of this scale. My previous research experience has mostly been through observing and helping around medical research rather than leading my own study. As xFilosofem moves further into actual research, I also want outside academic or domain-specific review where it is needed.

Our main track record is xFilosofem itself. We already have a working application, a functioning Russian learning domain, persistent learner history, provenance and revision handling, and 767 passing regression tests. We also have a research plan for the next stages, including prospective prediction, simple baseline comparisons, uncertainty, and later cross-domain testing with Medicine.

My partner has worked on several technical projects outside xFilosofem. These include indoor person tracking and trajectory mapping using image processing and homography, speaker diarization using both classical and neural methods, and a multimodal financial decision-support application using React, FastAPI, and AI-based image and intent analysis.

Neither of us is coming from a large company or an established lab. Most of what we have done has been built independently.

What are the most likely causes and outcomes if this project fails?

I do not think the main risk is that the idea suddenly reaches a complete dead end.

We already have a working system, an existing learning history, and a plan for how to test the next stage. The bigger risks are practical: having enough time, enough funding, and enough outside review to test things properly while I continue medical school.

There are also technical risks. The first models may turn out to be too complicated for the amount of data we have. If that happens, I would reduce the number of parameters and test simpler versions instead of trying to protect the original model.

The cross-domain part is very important to me, but I also do not expect Russian and Medicine to magically behave the same way. If the first cross-domain result is weak, I would look more closely at what can actually be shared and what needs to stay separate.

A failed test would not mean that I stop building xFilosofem. I care too much about the project for that. It would mean that one of my assumptions was wrong.

In a way, that is part of why the system is designed the way it is. I want the models and assumptions to be allowed to change without rewriting the history underneath them. If one direction fails, I would rather learn from that result and move in a better direction than force the original idea to look correct.

How much money have you raised in the last 12 months, and from where?

$0 in external funding. The project has been self-funded so far.