Back

2025

Backend

GunnerGPT

RAG system for Arsenal FC tactical knowledge - retrieves from ChromaDB with cited answers, full retrieval trace, and sub-2s latency.

GunnerGPT

Generic LLMs lack deep, specific tactical knowledge about Arsenal FC - they hallucinate player stats, misremember match outcomes, and can't cite sources.

Build a RAG system that answers Arsenal questions with grounded, cited evidence instead of guesses.

Curated a dataset of player profiles, tactical reports, and season archives, ingested into ChromaDB for cosine similarity retrieval. Built a RAG pipeline using Gemini 1.5 Flash that generates cited answers with an 'Intelligence Trace' UI showing every retrieval step, similarity scores, and source text.

0.85+

Avg Relevance

<2s

Latency

Architecture

Query -> Embedding -> ChromaDB -> Context + Prompt -> Gemini 1.5 -> Response

Stack

Next.jsFastAPIPythonLangChainChromaDBGoogle Gemini
Up Next

MeTTa AI Assistant