Spring 2025 · CUS 635 Mini-Projects

Team 1 — Benjamin Hanim, Peter Roumeliotis, Giulio Bardelli

Notebook	Goal	Key artifact
mini_project_1.ipynb	Scrape last-month news via News API & upload raw JSON to S3	137 source files in s3://cus635-spring2025/TEAM_1/sources/
mini_project_2.ipynb	Build 1 024-dim vectors & upsert to Pinecone	Namespace Team_1 in index cus635
mini_project_3.ipynb	LangChain / LangGraph RAG agent over Finance news	Colab demo that answers live queries

---

Quick-Start (Colab)

1. Open any notebook with the Open in Colab badge.
2. In Colab sidebar ▸ Secrets, add
   • OPENAI_API_KEY
   • PINECONE_API_KEY
3. Runtime ▸ Run all.
   If you only want to see the agent, jump straight to Mini-Project 3—vectors are already hosted.

ℹ️ Why do demo cells look blank on GitHub?
We cleared all widget outputs before committing to avoid
There was an error rendering your Notebook: the 'state' key is missing from 'metadata.widgets'.
After you press Run all, Cell 8 in mini_project_3 will display full answers.

---

Project 1 – News Scraper (mini_project_1.ipynb)

Step	Details
API	News API (sport, bitcoin, government queries; 30-day window)
Output	137 articles, grouped by source; each saved as <Source>.json
Storage	Uploaded via unsigned Boto3 to cus635-spring2025/TEAM_1/sources/

Rerun guide: set your own news_key, change TEAM, run the notebook—files will appear in your S3 folder.

---

Project 2 – Vector Ingestion (mini_project_2.ipynb)

1. Fetch the JSON files from S3.
2. Process each article → single text field (title + description + content).
3. Embed with dummy seeded 1 024-dim vectors (class exercise alignment).
4. Upsert to Pinecone → 137 vectors in cus635 / Team_1.
5. Sanity check sample queries (stock market, investment trends, …).

---

Project 3 – RAG Agent (mini_project_3.ipynb)

Layer	Tech	Notes
Embeddings	thenlper/gte-large (HF, 1 024-dim)	matches index dims
VectorStore	langchain_pinecone.PineconeVectorStore	Finance-only filter
RAG Core	ConversationalRetrievalChain	simple prompt template
Agent	LangGraph ReAct	single tool search_finance_news
LLM	GPT-3.5-Turbo	temp 0.2 answers, 0.1 reasoning

Demo Queries (Cell 8)