Cyber Infrastructure (CIS) · Sole developer · Live as PixelGPT
pixel-base
AI Knowledge & Support Assistant
An enterprise generative-AI assistant over internal support documentation, with an agent that operates JIRA through conversation.
Key metrics
- retrieval stores fused (vector, graph, SQL)
- 3
- inference backends balanced
- 3
What it does
- Built end to end: React frontend, Python/FastAPI backend, architecture through production.
- Hybrid RAG across ChromaDB vector search, a Neo4j knowledge graph with Graphiti temporal memory, and PostgreSQL.
- Conversational JIRA agent — create, read, update and delete tickets in natural language via LLM tool calling.
- Multi-model inference across OpenAI, self-hosted Ollama, and custom models on RunPod GPUs, balancing cost, latency and data privacy.
- Markdown-aware chunking and a custom reranker for retrieval precision on long technical documents.
- Dockerised and shipped through GitHub Actions CI/CD.
Stack
- Python
- FastAPI
- React
- ChromaDB
- Neo4j
- Graphiti
- Ollama
- OpenAI
- RunPod
- Docker