Rogersurf/trump_pulse_data

trump-signal# 🚀 TrumpPulse — MLOps Semester Project

👥 Group Members

• Chenghao Luo
• Suchanya Baiyam
• Rogerio Braunschweiger de Freitas Lima

---

📌 Project Overview

TrumpPulse is an MLOps-focused system designed to ingest, process, and analyze textual data from Trump Truth Social Dataset.

The primary goal is not model performance, but the design and implementation of a robust, reproducible, and deployable MLOps pipeline.

The system performs:

• 🧠 Sentiment Classification (positive / negative / neutral)
• ❓ Question Answering (QA) over Trump’s posts

This project aligns with Aalborg University’s MLOps requirements by emphasizing:

• End-to-end pipeline
• Deployment readiness
• Reproducibility
• Continuous or on-demand execution

---

🎯 Aim of the Project

The objective is to build a production-ready MLOps pipeline that includes:

• 📥 Data ingestion (initial + reusable)
• 🧹 Preprocessing
• 🧬 Feature representation (embeddings)
• 🤖 Model inference (sentiment + QA)
• 🌐 API-based deployment
• 🐳 Containerization (Docker)
• 🔁 Continuous or trigger-based execution

---

🏗️ System Architecture

        ┌────────────────────┐
        │ Hugging Face Dataset │
        └─────────┬──────────┘
                  ↓
        ┌────────────────────┐
        │ Data Ingestion     │
        └─────────┬──────────┘
                  ↓
        ┌────────────────────┐
        │ Preprocessing      │
        └─────────┬──────────┘
                  ↓
        ┌────────────────────┐
        │ Model Layer        │
        │ - Sentiment        │
        │ - QA (Embeddings)  │
        └─────────┬──────────┘
                  ↓
        ┌────────────────────┐
        │ FastAPI Service    │
        └─────────┬──────────┘
                  ↓
        ┌────────────────────┐
        │ Docker Container   │
        └────────────────────┘

---

⚙️ Pipeline Components

📥 Data Ingestion

• Data is loaded from Hugging Face
• Stored locally for reproducibility

🧹 Preprocessing

• Text cleaning
• Null filtering
• Basic normalization

🤖 Model Layer

Sentiment Analysis

• Pretrained transformer model

• Outputs:

  • Label (POSITIVE / NEGATIVE)
  • Confidence score

Question Answering (QA)

• Sentence embeddings using SentenceTransformers
• Semantic similarity search
• Returns most relevant Trump posts

---

🌐 API Deployment

The system is deployed using: 👉 FastAPI

Available Endpoints

Endpoint	Description
/	Health check
/predict_sentiment	Classify sentiment
/ask_question	Retrieve relevant answers

---

🐳 Containerization

The entire system is containerized using Docker:

• Ensures reproducibility
• Enables easy deployment
• Supports continuous operation

---

🔁 Execution Strategy

The system supports:

• ✅ API-based interaction (on-demand)
• ✅ Continuous execution via server runtime
• 🔜 Future: scheduled jobs (cron / GitHub Actions)

---

📦 Artifact Management

The project stores:

• 📁 Raw data
• 📁 Processed data
• 📁 Model outputs (predictions)

This supports:

• Reproducibility
• Debugging
• Evaluation

---

📊 Evaluation & Monitoring (Basic)

• Logging of predictions
• API health monitoring
• Model output inspection

Future improvements may include:

• Model versioning
• Drift detection
• Performance tracking

---

🧪 How to Run the Project

▶️ Local Run

pip install -r requirements.txt
uvicorn app.main:app --reload

Access: 👉 http://localhost:8000/docs

---

🐳 Docker Run

docker build -t trump-pulse .
docker run -p 8000:8000 trump-pulse

---

🚀 Future Work

• 📊 Market reaction modeling
• 📈 Time-series analysis
• 🧠 Advanced NLP models
• 🌍 Frontend dashboard (Streamlit)
• 🔄 Automated data pipelines

---

✅ Key Takeaways

• Focus on MLOps, not model perfection
• Demonstrates end-to-end pipeline
• Fully containerized and deployable
• Supports real-time interaction via API

---

📎 Repository

👉 (https://github.com/Rogersurf/trump-signal)

---

💡 Final Note

This project is designed to demonstrate a working, scalable, and reproducible MLOps system, aligned with academic requirements and real-world deployment practices.

---