Blog

• Teaching LM Studio to Browse the Internet When Answering Questions

  I really like LM Studio because it allows you to run AI models locally, preserving the privacy of your conversations with the AI. However, compared to commercial online models, LM Studio doesn’t support internet browsing “out of the box.” Those models can’t use up-to-date information from the Internet to answer questions.

• How to Create an AI Telegram Bot with Vector Memory on Qdrant

  The idea for this pet project came from my desire to build my own AI agent. I established minimal technical requirements for myself: the agent should have multiple states, be able to launch tools, and use RAG (Retrieval Augmented Generation) to search for answers.

• CBLT — A Safe, Fast, and Minimalistic Web Server in Rust Programming Language

  To learn a new programming language, I use the following approach. First, I read a tutorial for that programming language, which explains the syntax, idioms, philosophy, and principles of how the language works. After that, I write a small pet project in that programming language. In the pet project, I practice a bit with the new language, its standard libraries, and popular frameworks.

• rustsn — Open Source Project for Code Generation and Interaction with Existing Code Using LLM

  I’ve been working on a tool called rustsn, which allows generating, compiling, and testing code using LLMs (Large Language Models). Initially, the idea was to automate the process of writing small pieces of code — so-called snippets — for different programming languages based on user-provided explanations. This tool has since evolved and gained new features, such as generating complete code for applications and explaining existing code based on vector representations (embeddings).

• Metatron - Open Source library for generating reports in Rust language

  A year ago, the idea arose to rewrite the entire Java backend in Rust, which I have been developing and supporting for several years. I found all the analogues of libraries and frameworks from the Java world in the Rust ecosystem:

• Shiva — Open Source project in Rust for parsing and generating documents of any type

  The idea for the project came to me while working on a document search engine project. There is such a library like Apache Tika, written in Java, which can parse documents of various types. For my search engine to work, it must be able to extract text from documents different types (PDF, DOC, XLS, HTML, XML, JSON, etc.). I wrote the search engine itself in Rust. But, unfortunately, in the Rust world there is no library that can parse documents of all types.

• Simplified Rust example of training a neural network based on the Candle Framework by Hugging Face

  I was recently looking at examples of using neural networks from the Hugging Face Candle library and noticed that they are quite difficult to understand for people who are just starting to get acquainted with neural networks. Therefore, I decided to write the most simplified sample code in Rust, which demonstrates the training and use of the simplest neural network.