BautistaSobenko

Building AI-powered solutions for modern businesses

Software Engineer and Full Stack Developer with 4 years of experience specialized in translating business ideas into efficient, scalable software solutions.

Things I've built

RAG + LLMs

RAG-powered Travel Itinerary Generator

Retrieval-Augmented Generation system indexing 5,000+ activities with text-embedding-ada-002, stored in a vector database with metadata filters. Reduced hallucinations by 30% by grounding Claude in real activity data.

RAGpgvectorEmbeddingsClaude
AI + Computer Vision

Real-time military aircraft recognition

Computer Vision project using YOLOv7 for object detection and classification of military aircraft with CNNs.

PythonCNNsDeep LearningYOLOv7
AI + HR Tech

LLM-Powered ATS

Two-stage candidate screening system with algorithmic filtering followed by LLM analysis with structured JSON outputs. 75% reduction in screening time.

GPT-4EmbeddingsPythonJSON Mode
System Design

Instant messaging system

Chat application addressing Software Architecture quality attributes with server-client and P2P versions.

JavaSocketsP2PArchitecture
Multi-modal AI

OpenAI CLIP-based Image Selection Engine

Zero-shot image classification system leveraging CLIP embeddings to match images with text descriptions in a shared vector space. Automated aesthetic matching for destination pages without training custom classifiers.

OpenAI CLIPEmbeddingsPythonCosine Similarity
AI + Cloud

Document Processing with OCR + LLM

Web application for document upload and processing using AWS Textract for OCR, with LLM integration to improve accuracy and extract key data in clean, structured JSON format.

AWS TextractLLMFlaskPython
Computer Vision

Image Processing Pipeline

Automated content curation system using CLIP for semantic image-text matching across 20,000+ images. Multi-source integration with Google Search, Unsplash, and OpenAI Vision.

OpenAI CLIPOpenAI VisionPythonEmbeddings
Database Design

Laboratory security control database

Database solution for real-world security control using PostgreSQL with triggers, stored procedures, and AWS RDS.

PostgreSQLAWS RDSNormalizationER Diagram
AI + Automation

Ebook Processing Pipeline

Autonomous system that processed 5,000+ books using chunking strategies, sequential LLM processing, and DALL-E cover generation. 99.2% success rate with overnight batch processing.

GPT-4DALL-EPythonRate Limiting

A bit about me

Close to the problem

I don't just write code — I need to understand the why behind it. I estimate timelines, break down problems, propose technologies, and debate trade-offs with founders. I like being part of the decisions, not just executing them.

AI-native mindset

AI is not a buzzword in my workflow — it's how I operate. Claude Code, Cursor, LLM APIs in production. RAG pipelines, embeddings, prompt engineering — not because it's trendy, but because it lets me ship faster.

Full-stack, end to end

I own projects from idea to production — backend, frontend, infrastructure, deployment. I pick whatever tool solves the problem best, not whatever I'm most comfortable with. That's how I've always worked.