Android product engineering
Design and development for Android applications, utilities, games, and experimental consumer products.
Android · LLMs · AI systems
Ideas into intelligent software.
Mystery Box builds Android products, explores language-model research, and helps teams turn AI opportunities into working systems.
Product engineering
LLM agents & reasoning
Rapid AI prototyping
What we do
Three ways to build something useful.
From a focused mobile product to an experimental LLM system, the goal is the same: reduce ambiguity, build quickly, and make the result usable.
LLM & AI research
Research-driven prototypes around language models, agents, mathematical reasoning, state-space search, embeddings, and evaluation.
AI consultancy
Technical guidance and hands-on prototyping for teams deciding where AI fits, which approach to use, and how to turn it into a reliable workflow.
Selected Android work
Products built to be played, used, and explored.
A selection of public projects. More experiments and product work live across the GitHub profile.
01 / Logic game↗
Deduce It
A deduction-focused Android game built around compact logic challenges and puzzle solving.
02 / Learning tool↗
Interview Sprouts
An Android project focused on structured interview preparation and lightweight learning workflows.
03 / Interactive app↗
Mentalism
An interactive Android experiment exploring performance, perception, and playful user interaction.
04 / Experience app↗
Hypnosis
A compact Android experience project centered on guided interaction and focused presentation.
LLM research
Reasoning beyond the chat box.
Current interests include language-model agents, mathematical and theorem-oriented reasoning, exploration of large state spaces, representation learning, and systems that combine learned models with explicit search or verification.
Browse research repositories ↗
Reinforcement Math Theorems ↗
Experiments connecting learning systems with mathematical theorem-oriented tasks.
Neural Theorem Similarity ↗
Exploration of learned representations and similarity for theorem-related content.
Generative Modeling ↗
Earlier experimentation with generative neural-network methods.
About Mystery Box
A small studio for ambitious technical ideas.
Mystery Box is an independent software and AI practice spanning mobile products, machine learning research, and technical consulting.
The work sits between research and product engineering: understand the problem precisely, prototype the hard part early, measure what matters, and ship the simplest system that genuinely solves the problem.
Prototype earlyTest the difficult assumption before polishing the easy parts.
Measure realityUse evaluation and feedback rather than relying on impressive demos.
Build to shipPrefer practical systems that can survive outside a notebook.
Start a project
Have an app, AI workflow, or research problem?
For collaborations, consulting, prototypes, or technical discussions, reach out through GitHub and include a short description of what you are trying to build.
Open GitHub profile ↗