Skip to content
View RaNurbekov's full-sized avatar
  • Almaty, Kazakhstan

Block or report RaNurbekov

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
RaNurbekov/README.md

Hi, I'm Rashid 👋

ML Engineer | Fintech & Generative AI

Building production ML systems for financial services — from model training to Docker deployment

Profile Views Telegram Email


👨‍💻 About Me

I'm an ML Engineer focused on Fintech products and Generative AI. I specialize in the full development cycle: data analysis → model training & optimization → production deployment via FastAPI + Docker.

I don't just train models in Jupyter — I build reliable, interpretable and scalable ML pipelines ready for real business use.

Core expertise:

  • 🏦 Fintech ML — credit scoring, fraud detection (GNN + classical ML), PFM analytics
  • 🤖 LLM & Agents — Fine-Tuning (LoRA/PEFT), RAG, Voice AI, LangGraph ReAct Agents
  • ⚙️ MLOps — MLflow, Evidently AI, Kafka Streaming, A/B testing, Docker

🛠 Tech Stack

Machine Learning

Python PyTorch Scikit-Learn LightGBM HuggingFace PyG SHAP

LLM & Voice AI

LangChain LangGraph Qdrant LoRA Whisper Groq

Backend & MLOps

FastAPI Docker Redis Kafka MLflow Evidently Streamlit


💼 Projects

Fraud detection via transaction graph analysis using Graph Convolutional Networks

Транзакции представлены как граф: 590K узлов, 493K рёбер. GCN видит транзакцию в контексте её связей — если карта участвовала в подозрительных операциях, модель "заражает" соседей подозрением. Именно так работают антифрод системы Visa и Mastercard. AUC-ROC: 0.78, Fraud Recall: 62%.

🔗 API | Graph Visualizer

PyTorch Geometric GCNConv NetworkX FastAPI Streamlit Render


Hybrid real-time fraud detection microservice

Двухслойная архитектура: Redis Velocity Check (защита от брутфорса) + LightGBM (поведенческие паттерны). Встроенный A/B Testing Router распределяет трафик 80/20 между Champion и Challenger моделями.

🔗 Live API

LightGBM Redis Feature Store Docker Compose FastAPI A/B Testing Render


Full MLOps pipeline for bank credit scoring — from Kaggle dataset to deployed REST API

LightGBM + FastAPI. MLflow Model Registry — динамическая загрузка модели по RUN_ID. SHAP — топ-5 факторов решения. Evidently AI — мониторинг Data Drift. SQLite аудит-лог.

🔗 Live API

LightGBM MLflow SHAP Evidently AI FastAPI Docker Render


End-to-end AI Personal Finance Manager — localized for Kazakhstan 🇰🇿

NLP категоризация транзакций → Anomaly Detection → Prophet Forecast → Voice Assistant → PDF Reports → Telegram Bot. Симуляция KZ клиентов (Glovo, Kaspi, inDrive). Мультивалютность (8 валют).

🔗 Live Demo

Prophet Evidently AI Whisper Llama-3 Telegram Bot Streamlit


Real-time fraud detection streaming pipeline on Apache Kafka

Event-Driven пайплайн: Producer → Kafka Broker → Consumer Group → LightGBM ML API. Consumer вызывает fraud-detection-api для реального ML инференса вместо Rule-Based логики.

Apache Kafka Zookeeper Docker Compose Consumer Groups ML Integration


Autonomous Financial AI Agent with 5 tools (LangGraph ReAct)

LangGraph ReAct агент с инструментами: DuckDuckGo Search, Loan Calculator, Deposit Calculator, Currency Converter, Fraud Risk Checker. Сам решает какой инструмент вызвать.

🔗 Live Demo

LangGraph LangChain Llama-3.3-70B Tool Calling Groq Streamlit


Intelligent bank support assistant with semantic search over knowledge base (RAG)

Полноценная RAG-архитектура: документы → чанкинг → HuggingFace Embeddings → Qdrant → Llama 3. Hallucination Prevention: бот отвечает строго по документам.

🔗 Live Demo

Qdrant RAG LangChain Llama-3 HuggingFace Embeddings Streamlit


Full voice pipeline: Speech-to-Text → LLM → Text-to-Speech

Whisper Large v3Llama 3.3-70BgTTS. Память всего диалога. System Prompt оптимизирован под TTS: краткие ответы как живой оператор по телефону.

🔗 Live Demo

Whisper v3 Llama-3.3-70B gTTS Groq Conversation Memory Streamlit


Parameter-Efficient Fine-Tuning of LLM for bank support (PEFT/LoRA)

Fine-Tuning TinyLlama-1.1B — обучалось 1,126,400 параметров (0.10%). Cloud-to-Local Pipeline: Colab T4 GPU → CPU inference. Streamlit Chat UI.

🔗 Live Demo

TinyLlama LoRA PEFT SFTTrainer PyTorch HuggingFace


NLP system for "dirty" bank transaction categorization (PFM)

Fine-Tuning многоязычного DistilBERT на 5 банковских категорий. Устойчив к опечаткам, латинице и шумам POS-терминалов. Confidence Score UI.

DistilBERT PyTorch Hugging Face Fine-Tuning NLP Streamlit


Bank churn prediction — custom PyTorch MLP from scratch | ROC-AUC: 0.9685

Кастомный Training Loop на чистом PyTorch. BatchNorm + Dropout. Обучен на реальном датасете (10,127 клиентов). Бизнес-рекомендации по уровню риска.

🔗 Live Demo

PyTorch MLP BatchNorm ROC-AUC 0.97 Streamlit


Real-time object detection web application (YOLOv8)

Zero-Shot Inference на MS COCO (80 классов). Выбор модели от Nano до XLarge. Confidence slider. Plotly analytics dashboard.

🔗 Live Demo

YOLOv8 Ultralytics OpenCV Zero-Shot Plotly Streamlit


📊 GitHub Stats

RaNurbekov's GitHub Stats Top Languages

GitHub Streak


📫 Get in Touch

📧 Email nurbekovrashidjob@gmail.com
💬 Telegram @Ytyglika
📍 Location Almaty, Kazakhstan 🇰🇿

"Data without action is just numbers. I build systems that make decisions."

Pinned Loading

  1. credit-risk-api credit-risk-api Public

    Full ML Engineering cycle for bank credit scoring: Training → MLflow Tracking → FastAPI → SHAP Explainability → Drift Monitoring → Docker

    Python

  2. fraud-detection-api fraud-detection-api Public

    Hybrid fraud detection API: Redis Feature Store + LightGBM + A/B Testing Router. Live on Render.

    Python

  3. bank-llm-finetuning bank-llm-finetuning Public

    Parameter-Efficient Fine-Tuning of TinyLlama-1.1B for bank support LoRA adapters — training only 0.10% of parameters | Streamlit Chat UI | Cloud-to-Local Pipeline

    Python

  4. fraud-gnn fraud-gnn Public

    Graph Neural Network fraud detection on 590K transaction graph. GCN + FastAPI + Streamlit visualizer. IEEE-CIS dataset. AUC-ROC: 0.7807

    Python

  5. bank-ai-assistant bank-ai-assistant Public

    Intelligent bank support assistant with semantic search over knowledge base Qdrant Vector DB + HuggingFace Embeddings + Llama 3.3-70B | Hallucination Prevention

    Python

  6. pfm_ai_assistant pfm_ai_assistant Public

    End-to-end Personal Finance Manager: NLP categorization, anomaly detection, Prophet forecast, voice assistant, PDF reports & Telegram bot. Localized for Kazakhstan 🇰🇿

    Python