DAUZ / 26
PORTFOLIO / 202641.0°N / 72.3°E
NLP-focused developer · web · Telegram bots · Andijan, Uzbekistan

DOSTONBEKABDURAKHMONOV

I turn language problems into useful software — from NLP pipelines to web products and Telegram bots.

Dostonbek Abdurakhmonov in a dark studioNLP / WEB / TELEGRAM · ANDIJAN

Focused on practical NLP. Working knowledge of machine learning and computer vision.

Enter the workResume / PDF
NLPprimary direction
PYmain language
WEBproduct delivery
BOTSTelegram automation

GitHub portfolio / 01—02

Two focused NLP systems. Built from the tokenizer up.

Scroll through the technical case files

CASE01/ 02
NER PIPELINE / UZ · EN · RUMODEL: ACTIVE
MEN TOSHKENTDAGI OPERATORGA MUROJAAT QILDIM
LOCORGQUERY
TOKENS → EMBEDDING → BiLSTM → CRF → ENTITIES
NER PIPELINE / UZ · EN · RUMODEL: ACTIVE
MEN TOSHKENTDAGI OPERATORGA MUROJAAT QILDIM
LOCORGQUERY
TOKENS → EMBEDDING → BiLSTM → CRF → ENTITIES
01ENTITY RECOGNITION FROM SCRATCH

Uzbek Operator NER

NLP developer

A multilingual named-entity recognition system built from the ground up with a custom training pipeline and operator-domain entities.

The work
Dataset preparationSequence model designTraining and evaluation
Outcome
An open-source multilingual NER prototype with a reproducible pipeline.
Technology
Python · PyTorch · BiLSTM-CRF · FastText
Open project
QUERY / UZBEK OPERATORTOP-K / 05

Savolimga aniq javob top.

010.92BILLING POLICY / OPERATOR020.87SUPPORT GUIDE / UZBEK030.81PRODUCT CONTEXT / FAQ
BPE → BERT 50M → SIMCSE → FAISS → RESPONSE
02RETRIEVAL SYSTEM FROM SCRATCH

Uzbek Operator RAG

NLP developer

An end-to-end retrieval-augmented generation pipeline with custom BPE, a BERT-style transformer and hybrid semantic retrieval.

The work
Tokenizer and encoderContrastive trainingHybrid retrieval pipeline
Outcome
An open-source, end-to-end retrieval prototype built for experimentation and extension.
Technology
Python · PyTorch · BPE · SimCSE · FAISS
Open project
Visit GitHub portfolio

Developer profile / capability map

Language first. Products always.

My strongest direction is NLP. I also build production-minded web interfaces and Telegram automations, with supporting knowledge in ML and computer vision.

PRIMARYNLP development

Tokenization, NER, retrieval and language-focused model pipelines.

DELIVERYWeb development

Responsive products and interfaces with TypeScript and Next.js.

AUTOMATIONTelegram bots

Python bots for publishing, workflows, forms and user interactions.

SUPPORTINGML & Computer Vision

Practical foundation for classification, embeddings and visual search.

Current stackLOW-RESOURCE NLP×PYTHON×FASTAPI×NEXT.JS×TELEGRAM BOTS×PYTORCH

Proof / competitions

Scoreboard, not decoration.

Ranks and awards that sharpened the practice.

011st place

Sharda University Olympiad

Regional / 100% scholarship for 4 years
021st & 6th place

Town & Region Informatics Olympiad

Town & Region
0373rd

IOI (International Olympiad in Informatics)

Republic of Uzbekistan
0413th

IOAI (International Olympiad in Artificial Intelligence)

Republic of Uzbekistan
05750th

Kaggle AIMO-3

Global / Out of 4,000 participants
06Top 1000

Kaggle Stanford RNA 3D Folding Part 2

Global / First Kaggle competition
07Participant

Kaggle Tregegeist Hackathon

Global

Learning archive

Learning with a direction.

Selected Kaggle credentials from a wider learning path: Python, deep learning and applied machine learning — the foundations behind my NLP work.

Python certificate
01 / Python
Intro to Deep Learning certificate
02 / Intro to Deep Learning
Intermediate Machine Learning certificate
03 / Intermediate Machine Learning

Open to IT roles / internships / projects

ANDIJAN / UZBEKISTAN

Let’s build something useful.

I use this portfolio as a complete professional snapshot. Contact me about NLP, Python, web development or Telegram bot opportunities.

TelegramGitHubHugging Face