Ask the AI Chatbot
Ask anything about 35+ Trump rally speeches — powered by a 10-stage RAG pipeline built from scratch: query rewriting, hybrid semantic search, cross-encoder re-ranking, grounded answer generation, and multi-factor confidence scoring. Three ML models under the hood. Zero LangChain.
Quick test on these topics:
📚 Supporting Context — click to expand retrieved passages
🎭 Sentiment Analysis
Drop in any text and get a three-model breakdown — FinBERT scores the sentiment, RoBERTa maps the emotions, and an LLM ties it together with a plain-English interpretation. Speeches, news articles, tweets, whatever.
🏷️ Topic Extraction
Drop in any substantial text and get it broken down into meaningful themes — keywords grouped by semantic similarity, real examples pulled from the source with highlighting, and an LLM summary tying it all together. Under the hood: DBSCAN clustering with sentence-transformers.
📊 The Dataset
35 Trump rally speeches (2019–2020) chunked into ~2,354 semantic segments, embedded as 768-dimensional MPNet vectors, and indexed in ChromaDB — the knowledge base the RAG chatbot searches. Load the stats to see speech lengths, word counts, and corpus-wide breakdowns.
Loading statistics...