DeskHead is an innovative AI-powered trading platform developed in collaboration with MIT-sponsored research. It features AI agents such as Risk Analyst AI, Researcher AI, and Trader AI to assist traders with real-time predictions, risk assessments, and market research. By addressing core challenges like missed price spikes, scattered historical data, and trader skepticism, DeskHead has transformed the trading experience, empowering traders to make faster and more reliable decisions.

1. Real-Time Data Pipeline
   Built with Kafka to process market feeds and news data instantly, reducing latency significantly.

2. Centralized Historical Data Storage
   MongoDB integration ensures all historical stock events are systematically stored, enabling easy analysis.

3. Spike Prediction Model
   Leveraged DBSCAN clustering to identify patterns and predict price spikes with 40% higher accuracy.

4. Explainable Predictions
   Historical precedents are linked to predictions, fostering trader trust and confidence.

5. Lightning-Fast Insights
   Predictions are delivered in under 2 seconds, enabling traders to act on market movements without hesitation.

1. Missed Opportunities
   Price spikes often went unnoticed or were flagged too late for action.

2. Fragmented Historical Data
   Disorganized data storage made it difficult to recognize actionable patterns.

3. Skeptical Traders
   Lack of trust in predictions due to insufficient explainability and accuracy.

• Built a robust data pipeline for ingesting and organizing historical and real-time data.
• Developed a scalable prediction model to detect both expected and unexpected price spikes.

• Implemented DBSCAN clustering for pattern recognition.
• Integrated AI agents to dynamically recalibrate predictions based on live data.

How it Works:

1. The user (Desk Head) triggers AI agents based on prompts or events.
2. Agents reference custom tools and retrieve data, synthesizing complex information into actionable insights.
3. Alerts, updates, and relevant links are provided to users in real-time.

1. Data Overload
   Filtered thousands of daily inputs from news articles, social media, and market feeds to prioritize relevance.

2. False Positives
   Improved model reliability with trader feedback and explainability features.

• 40% Improvement in Spike Prediction Accuracy
  Enhanced traders' ability to identify unexpected market movements.

• Centralized Historical Database
  Organized data storage with MongoDB for seamless analysis and learning.

• Sub-2-Second Prediction Speeds
  Real-time insights empower traders to act without delays.

• Restored Trader Confidence
  Transparent predictions linked to historical patterns increased trust in the system.

• Data Storage: MongoDB, SQLite
• Real-Time Processing: Kafka
• Machine Learning: DBSCAN Clustering, YFinance API, PolygonAPI, FinnHub Utils.
• Programming Languages: Python
• Visualization: Custom-built interactive interfaces

• Python 3.7 or later
• MongoDB
• Kafka

cd Finrobot
pip install -r requirements.txt

#Run Command
cd api
python app.py