Projects
A collection of projects showcasing my development skills and experience
Last Z Vector RAG Bot
AI-powered Poe chatbot that analyzes Last Z: Survival Shooter screenshots and provides strategic gameplay advice using semantic vector search. Identifies heroes, buildings, and resources from images to deliver personalized tier lists, upgrade recommendations, and progression strategies. Features dynamic RAG with 150+ game knowledge items, disk-based embedding cache, and anti-hallucination guardrails. Powered by ChatGPT-5 chat model for intelligent conversational analysis. Live on Poe.com with 300+ active users relying on it for strategic gameplay insights.
Python, FastAPI, Sentence-Transformers, Vector Embeddings, Poe API, Render, OpenAI, ChatGPT-5
Intelligent Context Agent
Two-tier agent architecture that manages long-term conversation context in Milvus vector database, using Ollama to intelligently optimize context passed to ChatGPT. Proactively predicts memory needs before LLM calls, maintaining efficient context windows and eliminating catastrophic context dumping. Features semantic memory retrieval, token-aware context assembly, and single-shot optimized responses across multiple conversation sessions. Deployed with Docker orchestration (Milvus, Ollama, FastAPI).
Python, FastAPI, Ollama, Milvus, LangChain, OpenAI, Docker, MCP