### AI Engineer | Data Science Student | Python Developer
I'm a Software Engineering graduate from Tashkent University of Information Technologies (TUIT) with a strong interest in Artificial Intelligence, Machine Learning, and modern backend development.
I'm passionate about building intelligent software powered by data and modern AI.
My current focus is on Machine Learning, Retrieval-Augmented Generation (RAG), Large Language Models (LLMs), Natural Language Processing (NLP), and AI-powered applications.
I enjoy building intelligent applications—from data processing and vector databases to backend APIs, Telegram bots, and deploying production-ready solutions.
Beyond AI, I work with Python, FastAPI, SQL, Linux, Docker, PostgreSQL, Git, and cloud technologies while continuously expanding my knowledge of modern Data Science and Machine Learning.
I believe the best way to learn is by building real-world projects, so most of my time is spent experimenting with new ideas and turning them into practical software.
QarzFlow — Telegram Expense Platform
Expense & debt tracking platform with a Telegram Bot, backend APIs, and a web dashboard — live in production.
| Stack | FastAPI, Telegram Bot API, PostgreSQL, Docker |
| Scale | Telegram bot + web dashboard + backend APIs over one shared platform |
| Performance | Async FastAPI backend with PostgreSQL persistence |
| Security | Containerized deployment with per-user data isolation |
| Impact | Live product at qarzflow.uz serving real users |
| Repository | qarzflow.uz · @qarzflowbot |
JOA — AI Coding Agent
An autonomous AI coding agent that plans, edits, and verifies code changes through a local-first pipeline.
| Stack | Python, Ollama, local LLMs, CLI/REPL interface |
| Scale | Multi-file codebase operations with incremental context management |
| Performance | Runs fully on consumer CPU hardware — zero GPU dependency |
| Security | Local-first by design: no source code leaves the machine |
| Impact | Daily-driver coding assistant with tracked clone growth and public analytics |
| Repository | TeamLider9141/JOA-AI-CODING-AGENT |
Built to prove that a capable coding agent does not require cloud inference. The agent handles model selection, LaTeX-to-Unicode answer cleaning, uncertainty nudging, and ships with clone-statistics automation baked into CI.
System LLM — Local Retrieval-Core Coding Assistant
A retrieval-augmented coding assistant with a hand-written retrieval core — no framework dependency — engineered for strict hardware budgets.
| Stack | Python, Ollama, qwen2.5-coder 7B, nomic-embed-text, custom vector store |
| Scale | Indexes real-world codebases in place — no corpus duplication |
| Performance | Tuned for CPU-only inference on 16 GB RAM (Ryzen 5800U class hardware) |
| Security | Fully offline operation; the index never touches a network |
| Impact | Replaces a LlamaIndex-based prototype with a leaner, fully understood core |
| Repository | Private — in active development |
The deliberate rewrite from framework-glue to first-principles retrieval: custom chunking, embedding, and ranking layers built to be read, measured, and tuned line by line.
Uzbek AI — Native-Language AI Tooling
AI tooling and assistant workflows built for the Uzbek language — a low-resource NLP environment where off-the-shelf models underperform.
| Stack | Python, LLM prompt pipelines, custom skill/workflow definitions |
| Scale | Language-specific terminology handling across technical domains |
| Performance | Optimized prompt and context strategies for smaller local models |
| Security | Local execution; no user text sent to third-party services |
| Impact | Makes serious AI-assisted engineering workflows usable in Uzbek |
| Repository | github.com/TeamLider9141 |
Bridges the gap between English-centric AI tooling and native-language engineering practice — from terminology mapping to full bilingual technical workflows.
AI Phone Recommender Bot
AI-powered Telegram bot that recommends phones based on user needs and budget.
| Stack | Python, Telegram Bot API, AI-driven recommendation logic |
| Scale | Conversational recommendation flow over a phone catalog |
| Performance | Lightweight bot backend, runs on modest hardware |
| Security | No personal data retention beyond the chat session |
| Impact | Practical AI assistant for a real purchase decision |
| Repository | TeamLider9141/AI_phone_recommender_bot |
learning:
- Machine Learning, MLOps, AI Agents, and Vector DBs
- Advanced retrieval architectures and embedding-space evaluation
- Distributed systems patterns for real-time multiplayer backends
building:
- QarzFlow — Telegram expense & debt platform (qarzflow.uz)
- System LLM — framework-free local RAG coding assistant
- JOA — autonomous local AI coding agent
exploring:
- Small-model agentic performance on CPU-only hardware
- Low-resource NLP for the Uzbek language
open_to:
- Software Engineer / AI Engineer /
- Open source collaboration on local-first AI tooling