Skip to content

KremlevLev/KremlevLev

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

37 Commits
 
 
 
 

Repository files navigation

𝔏𝔢𝔳 𝔎𝔯𝔢𝔪𝔩𝔢𝔳

"𝔅𝔲𝔦𝔩𝔡𝔦𝔫𝔤 𝔱𝔥𝔦𝔫𝔤𝔰 𝔣𝔯𝔬𝔪 𝔰𝔠𝔯𝔞𝔱𝔠𝔥 𝔲𝔫𝔱𝔦𝔩 𝔱𝔥𝔢𝔶 𝔴𝔬𝔯𝔨. 𝔚𝔢 𝔞𝔯𝔢 𝔫𝔬𝔱 𝔶𝔬𝔲𝔯 𝔨𝔦𝔫𝔡𝔰."

ML Engineer focused on LLM systems, AI agents and deep learning.
MIREA — Artificial Intelligence & Machine Learning. Hackathon finalist. Building transformers on JAX for fun.


⚔️ About

I build end-to-end ML systems — from raw data and feature engineering to production inference and deployment. Currently deep into LLM internals, agent architectures and RL theory.


𝔓𝔯𝔬𝔧𝔢𝔠𝔱𝔰

🏦 High-Load RAG Pipeline — Alfa-Bank & MIPT Hackathon

Production RAG system (~7000 queries) optimized for BERTScore-Recall-L

  • FAISS (bge-m3) + BM25 + RRF · cross-encoder reranking
  • vLLM continuous batching · 2×T4 GPU-aware architecture
  • Latency reduced by ~25%
  • → github

🛡️ PII Leak Detection — ARENA DATA (Top 6 / 25 teams)

ML pipeline for personal data leak detection

  • 3-tier architecture: rule-engine → classification → scoring
  • TF-IDF + LogReg · CatBoost with contextual scoring
  • 35% FP reduction
  • → github

🎯 Retail Video OCR Pipeline — Lenta CV Hackathon

Production CV system for price tag extraction from video stream

  • YOLOv8 + custom IoU tracker · PaddleOCR → EasyOCR cascade
  • Throughput: 7 → 19 FPS · OCR inferences reduced by 22%
  • → github

📈 Retail Turnover Forecasting — X5 Tech Gradient (Final)

Predictive model for store RTO forecasting · 90.45 score

  • Time-series feature engineering · gradient boosting · overfitting control
  • → github

⚙️ BPE Tokenizer from scratch

Byte-level BPE tokenizer · GPT-2 regex · O(n log n) complexity

  • Built for deep understanding of LLM tokenization internals
  • → github

🔬 LLM from Scratch — JAX

Transformer implementation: Self-Attention · Multi-Head · MoE · MLA

  • Researching memory efficiency and inference optimization
  • RAG experiments on custom architecture — ongoing

𝔖𝔱𝔞𝔠𝔨

stack = {
    "ml / dl":   ["PyTorch", "JAX", "CatBoost", "XGBoost", "Scikit-learn"],
    "llm / nlp": ["vLLM", "LangChain", "LangGraph", "FAISS", "BM25"],
    "infra":     ["Docker", "FastAPI", "SQL", "Git"],
    }

𝔉𝔦𝔫𝔡 𝔪𝔢

GitHub Streak

About

No description, website, or topics provided.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

No releases published

Packages

 
 
 

Contributors