Drag a sticker or use the arrow keys to turn it. Home restores the front view.
Speech, keywords and accents.
Experiments2020-2021
Keyword spotting: an example from my PyTorch implementation.QuartzNet architecture. Kriman et al., 2019, Fig. 1.
I implemented QuartzNet speech recognition from scratch in PyTorch, alongside attention-based keyword spotting and an accent-classification team project.
I tested whether expression trees helped Transformers integrate expressions and solve differential equations. The idea did not beat the baseline; the evaluation anomalies were the interesting result.
My degree was in Applied Mathematics and Computer Science. Giving the model explicit expression-tree structure did not improve the baseline overall. After removing structurally duplicated test examples, the integration baseline fell from 87.70% to 80.17%.
Timur is surprised that there is a so-called "summarization_bot". It needs to be added to the chat or sent messages from the chat.
An actual chat and the bot's summary.
300k+Hugging Face model downloads1,000+daily requests, historically$100+/monthhistorical revenue
I taught three school pupils self-attention, then we implemented Transformers in PyTorch and built a Russian chat-summary bot. Keeping it going became my first experience managing a team, subscriptions and server bills.
From the first lesson to closure / model, sources & note
We started before ChatGPT. I kept the project going after the original intensive program, designing features around user requests. Our teaching implementation was pure PyTorch; the deployed summarization model was a fine-tuned mBART, which remains available.
The example shows a chat about using the bot: adding it to a group or forwarding messages. The project became financially self-sustaining, but that didn't last. When it closed, subscriber support covered less than half of what we needed to break even.
NLP engineering2022-2023LLM memory, summarization and retrieval
I built LLM memory and retrieval systems for EVA AI. Integrating conversation summaries gave the chatbot 3x longer memory and increased ARPU by 9%.
~500kMAU / ML team of <5$25k/daytotal app revenue at the time3xlonger memory+9%ARPU from summarization
EVA AI today: official App Store image, October 2026.
One small fix: Llama 2 could reply in other languages, but our English-only classifier missed sexting that should have been billed. A multilingual encoder and a head trained on existing English labels fixed it in a day.
The engineering behind the numbers / sources
The diagram illustrates that multilingual fix; its sample languages are examples. The main work was on memory, retrieval and dialogue features:
Dialogue summarization
I integrated summaries to retain more conversation: 3x longer memory and a 9% increase in ARPU.
Questions about yesterday
Generating questions about the previous day's dialogue increased ARPU by 8%.
Fact extraction and retrieval
I built a pipeline using a custom LLM and vector database for a product serving roughly half a million monthly active users.
Automated error classification
I automated collection, labeling, training and deployment. Unethical responses fell by 35%, as measured by OpenAI's safety classifier.
A smaller classifier
A distilled, quantized BERT-like classifier replaced rules and improved quality at roughly the same throughput.
These are historical results. Revenue is for the whole app. The ARPU changes describe separate features.
I built a Mercedes EV sales assistant with Denis and Anastasiia Chaikova using specialist agents, retrieval and a supervisor. Second among more than 80 teams.
A review connecting HiPPO, linear recurrent models, and attention/state-space duality.
3D reconstruction from point clouds.
Poster2025 / TUM ML43D
With our Convolutional Occupancy Networks poster.
Noisy point-cloud reconstruction with Anastasiia Chaikova, Omar Al-Dabooni and Yi Wang: sinusoidal activations, positional encodings and semantic conditioning.
My TUM master's thesis with JetBrains asked whether a coding agent could carry less context between steps. I used an encoder to compress tool output into fixed-size memory embeddings for the language model. The compressed memory retained answers on single-shot tasks, but agents solved fewer software-engineering issues.
I studied implicit context condensation for local software-engineering agents: compressing the context an agent carries between steps. The later paper examines whether reconstruction errors compound across steps and why the compressed context might lose information needed later.
The poster photograph is from a JetBrains Research gathering, shared in October 2025.
This poster records partial research on choosing a model for each step of a software-engineering agent. Early prefixes provided weak routing signals; training raised problems with exploration and the cost-quality tradeoff.
Qwen3.5-4B / 4k training cap / AIME 2024/25 avg@3 / against the base model at each test budget.
The initial idea was mine: what does a fixed training budget teach a reasoning model? I noticed that giving trained models more tokens could put them behind the baseline. We investigated that reversal.
Setup, limitations, poster & sources
Training improved answers near the training cap, but could weaken the ability to benefit from more test-time tokens. Evaluating only at the cap would have hidden the reversal shown above.
The study covers three small model families, with one training seed per condition. Shorter responses remained useful on some transfer tasks. Keeping the gains without losing high-budget performance remains an open problem.
Bachelor thesis testing structural Transformer modifications for integration and ordinary differential equations; studied changes did not improve the baseline.
