← Home
MY PROJECTS

Real projects built from scratch

From LLM evaluation systems to mobile AI apps — every project solves real problems.

🏆 Flagship Project📊 50 test cases🛡️ 4-layer protection

🚀 LLM Evaluation System & Culinary App

A two-component project combining an LLM Evaluation Framework (Python, 19 files, 50 test cases) with the ZjemTo mobile app (React Native, 29 screens). The system detects AI hallucinations, supports GPT-4o, Claude, and the Polish Bielik models. Includes a multi-layered hallucination protection system: Judge System, Bielik validation, Polish Guard, and a local database of ~48 recipes.

PythonReact NativeGPT-4oClaudeBielikOllama
View project
24/7
monitoring

Competitor Spy

An AI and LLM-powered competitive intelligence platform. A system for monitoring competitor websites, web scraping (Jina AI), and SEO analysis. Detects content changes and generates strategic insights using language models. Designed as a SaaS application with architecture prepared for paid plans, subscriptions, and premium features.

GPT-4Jina AIFastAPIWeb ScrapingSEO
RAG
chatbot

AI Portfolio / Interactive CV

A web application built with Next.js and TypeScript, serving as an interactive CV. Features a RAG-based chatbot that answers questions about projects and skills based on defined context, with hallucination mitigation. The project includes multilingual support (PL/EN/DE), user behavior analytics, and architecture ready for further product development.

Next.jsRAGGPT-4o-miniTypeScriptUpstash Vector
🔨 Currently building
<50ms
alerts

SightNarrator - AI for the Visually Impaired

A mobile app for visually impaired users with a two-tier AI architecture. Safety Layer (local, <50ms): detects obstacles, stairs, vehicles; instant voice alerts; works offline. Narrative Layer (API, ~2s): describes scenes with context; various narration styles.

React NativeClaude VisionYOLOML KitExpo

Production systems

AI systems solving real problems in production

Document Q&A System with RAG

40%
fewer hallucinations

A production system processing 10k+ documents with hybrid search.

LangChainPineconeFastAPI

LLM Evaluation Framework

60%
faster reviews

Automated evaluation pipelines with custom metrics.

PythonOpenAIPytest

Cost-Optimized Chat Interface

50%
cost reduction

An intelligent caching layer reducing API costs, serving 1000+ users daily.

Next.jsRedisOpenAI

LLM Systems Architecture

1000+
daily users

Complete LLM systems: from data ingestion through reasoning and evaluation to monitoring.

FastAPIAsyncTracingMonitoring
Let's discuss your project
© 2026 Jakub Prejzner
System v2.0 | 2026