Tech Trends News Agent

An AI-powered news analysis system that retrieves technology articles, enables semantic search, generates conversational answers using a local LLM, and visualizes trending topics.

Colab Link

https://colab.research.google.com/drive/1BVgYnpfCmaCokEHSt9dL47DjziTT8nn0?usp=sharing

Video link

https://drive.google.com/file/d/1YYscs_qC2-t1McnWBjY0JgdyR-45pDfl/view?usp=sharing

Quick Start

1. Click the .ipynb notebook file in this repository
2. Click Open in Colab at the top
3. Set runtime to GPU (Runtime → Change runtime type → T4 GPU)
4. Run all cells (Runtime → Run all)
5. Click the ngrok URL when it appears to open the app

---

Setup and Run Instructions

1. Prerequisites

• Google account (for Colab)
• That's it! Everything else runs in the cloud.

---

2. Run in Google Colab

1. Click the notebook file (Copy_of_12_4_New_AI_project.ipynb) above
2. Click Open in Colab button at the top of the file
3. Go to Runtime → Change runtime type → Select GPU (T4)
4. Run all cells in order (Runtime → Run all)
5. Wait for the ngrok URL to appear (~60 seconds for model loading)
6. Click the ngrok URL to open the Streamlit app

---

3. API Key

You do not need to create your own NewsAPI key. A working API key is already included in the project configuration. The agent will automatically use it when retrieving articles.

---

4. Project Structure

tech-trends-agent/
├── app.py                 # Streamlit web interface
├── fetch_news.py          # NewsAPI article retrieval
├── preprocess.py          # NLTK tokenization and preprocessing
├── search_articles.py     # TF-IDF search and topic extraction
├── llm_interface.py       # Qwen2.5-1.5B model wrapper
├── data/
│   ├── raw/               # Raw JSON from NewsAPI
│   └── processed/         # Preprocessed article corpus

---

5. Using the App

Chat Tab

1. Click "Fetch Articles" in the sidebar to load latest technology news
2. Type a question (e.g., "What are the latest AI trends?") or click a suggestion
3. View the LLM-generated answer with expandable source citations

Topic Visualization Tab

1. After fetching articles, switch to the "Topic Visualization" tab
2. View bar chart of top 5 discovered topics
3. Expand each topic to see key terms, sample articles, and sub-trends

---

6. Output

• Chat Responses: Contextual answers displayed in the chat interface with source links
• Topic Clusters: Automatically extracted from article corpus using TF-IDF and co-occurrence analysis
• Sub-Trends: Relevance-scored sub-topics within each main topic cluster
• Metrics: Total articles, categorized count, and coverage percentage

---

7. Notes for Graders

• Easiest path: Run the Colab notebook with GPU runtime enabled
• The API key is pre-configured — no setup required for NewsAPI
• Model loading takes ~60 seconds on first run (downloading ~3GB)
• Once loaded, queries are answered in 2-5 seconds
• If NewsAPI rate limits are hit, previously fetched articles are cached in data/processed/
• The app automatically loads cached articles if available

To Test:

1. Run all cells in the notebook
2. Click "Fetch Articles" to retrieve latest news
3. Ask questions in the chat interface
4. Check the Topic Visualization tab for trend analysis

---

8. Technical Details

• LLM: Qwen2.5-1.5B-Instruct
• Search: TF-IDF with cosine similarity
• Preprocessing: NLTK (tokenize, stopwords, lemmatize)
• Topic Extraction: Co-occurrence clustering
• Frontend: Streamlit
• Data Source: NewsAPI (technology headlines)
• GPU Support: CUDA (float16 precision)

---

9. Troubleshooting

• "No articles loaded": Click "Fetch Articles" in sidebar
• Model loading slow: Ensure GPU runtime is enabled in Colab
• ngrok URL not working: Re-run the last cell to generate new tunnel
• NewsAPI error: Rate limit reached; wait or use cached data
• CUDA out of memory: Restart runtime and try again