Hi! I'm Sashko's AI - we can talk about my work, projects, or background.
Projects
Scene-Creation Assistant
A production AI system that converts natural language chat into valid JSON automation scenes for Shelly smart home devices. Built with RAG retrieval over JSON examples, validation layers, prompt engineering, and a backoffice dashboard. Started as an Alexa skill demo and evolved into a full chat-based feature integrated into the client platform. Ran in production for several months and is still active in staging.
Finstat Data
A data extraction and reporting pipeline for processing business documents and financial data. Combined OCR tools with OpenAI API to extract, validate, and transform information from PDFs and scanned images. The full pipeline covered image preprocessing, OCR extraction, validation, data processing, storage, and reporting output.
Shelly-Scripts
Embedded JavaScript scripts running directly on Shelly smart home devices, enabling local automation without depending on the cloud. Designed around an offline-first paradigm so devices continue operating even when internet connectivity is lost. Devices communicated via Bluetooth and Wi-Fi, and dozens of reusable scenario templates were created for end users.
Email Agent — 10-K Financial Data Extraction GitHub ↗
A production-style AI pipeline for extracting structured financial data from SEC 10-K reports using RAG, OpenAI, and rule-based validation. Includes FAISS-based semantic retrieval over a 9-document knowledge base, deterministic validators for accounting checks, MongoDB storage, Gmail API email polling, and an HTML review dashboard for accepted, needs_review, and failed documents.
askSashkoAi GitHub ↗
This website — a RAG-powered AI chatbot that speaks as me in first person. Built with FastAPI, OpenAI GPT-4o, FAISS, and a plain HTML/JS frontend. Includes rate limiting, session tracking, daily cost cap, email notifications, PostgreSQL, Alembic migrations, and deployment on Railway with Cloudflare DNS.
Fishbook GitHub ↗
A social platform for hobby fishermen built as my SoftUni diploma project. Users can share photos, like and comment on posts, and find information about lakes and weather conditions. Built entirely without AI coding assistants — everything was planned, implemented, and debugged manually, which made it a strong foundation for understanding full-stack development.
WeatherApp GitHub ↗
A multi-interface weather system built to demonstrate how the same core logic can be reused across different technologies. Includes a console app, a Tkinter desktop GUI, a Flask web app, and a Django web app with database support for comparing weather data between cities. Uses the OpenWeather API for real-time data.
Sashko Milushev
Python Developer — Backend & AI Solutions
I build reliable, scalable backend systems and AI-powered applications. My stack is Python-first — FastAPI, Django, SQLAlchemy — combined with modern AI tools like OpenAI, LangChain, and vector databases.
I care about clean code, good architecture, and building things that actually work. Whether it's a REST API, a RAG pipeline, or a full-stack solution — I like to understand the problem deeply before writing the first line.
Get in touch
Have a project idea, a job opportunity, or just want to say hi? Send me a message.