📨 Ask Enron – LLM Question Answering over the Enron Email Corpus

Live demo (Hugging Face Space): https://huggingface.co/spaces/singhalamaan116/Ask-Enron

Ask questions in natural language and get answers based on real Enron emails.
Under the hood, the app performs semantic search over the Enron email dataset and uses a small language model to generate answers from the retrieved emails.

---

🔧 Tech Stack

• Dataset: corbt/enron-emails
• Embeddings: sentence-transformers/all-MiniLM-L6-v2
• LLM: google/flan-t5-small
• UI: Gradio (Blocks)
• Infra: Hugging Face Spaces (CPU)

---

🚀 How it works

1. User enters a question (e.g., “What trips were people planning in August 2000?”).
2. The question is embedded with a sentence-transformers model.
3. We compute cosine similarity with precomputed embeddings of ~20k Enron emails.
4. The top-k emails are concatenated into a context.
5. A Flan-T5 model reads the context and generates an answer.
6. The app shows:
   • The answer
   • The exact emails used as context

---

📁 Project structure

app.py          # Gradio app (Spaces entry point)
requirements.txt
README.md