• 

  Sam Altman X Dashboard 1

• 

  Home Search

• 

  Sam Altman X Dashboard 2

Harkonnen

Harkonnen: Finnish for “bull-ish” / “bull-like” | colloquial for optimistic sentiment on stock price.

Project Description:

The problem:

Markets are increasingly influenced by social media, but there's no transparent, systematic way to measure or simulate that impact.

• Influential figures — politicians, CEOs, financial influencers—can move billions with a single post.
• Journalists and regulators — lack the tooling and quantitative support in exposing media market manipulation. Journalists can only call out the correlation between posts and market changes, but never draw conclusions.
• Established trading platforms — shy away from calling out narrative and influence.
• Level the Playing Field — Democratize the awareness of social-media driven price changes.

The Application of Harkonnen

Process social media posts from prominent social figures (politicians, CEOs, celebrities, etc) → evaluate sentiment and market relevancy → simulate trades.

Primary Features

1. Price Change Analysis

   • Given select tickers, ETFS, and a time-frame(1 day, 7 days), how did stocks prices change?
   • Fetch financial data from yahoo financial API (hackathon friendly).
     • yfinance ← python library
   • Influence Score:

   $$ \frac{\text{true positives + true negatives}}{\text{total predicitions}} $$

- Not indicative of model accuracy, rather the influence of the influencer

1. Front End Dashboard
   • Select/Search for politicians on social media platforms (X, Truth Social) and trigger backend data processing.
   • Create a social media like dashboard where posts are printed along with the price change analysis. (Allow options to change time frames)
   • Bonus:
     1. Create a leaderboards page
     2. Create a page describing our trading methods and simulations including our NLP pipeline for transparency.
     3. Documentation for our open source API.

2. NLP Pipeline For Sentiment Recognition and Semantic Market Influence

   Pipeline workflow:

   1. Scraping: Fetch and scope raw text social media posts for a given timeframe.
   2. Preprocessing: Text normalization + simple Regex clean.
   3. Sentiment Evaluation: Detect positive/negative influence given natural language.
   4. Entity Extraction:
      • Comb the post for any ticker’s/companies/products using an Named Entity Recognition (NER) dictionary.
      • Apply Fuzzy Search (String Approximation) on a simple data set of tickers (.csv)
   5. Optional: Implement retrieval augmented generation(RAG) to perform semantic based search. (”Tariffs are being applied to exports to metal and minerals” → “Texas Instruments”)
      • Would require creating an embedding space manually into SQLite.

3. REST API

   • Create needed endpoints for the frontend dashboard.
   • Keep good visualizations/data for demo.
   • Harkonnen doubles as an open-source financial nlp API for social media posts. So creating simple endpoints for API clients and documenting them is a bonus.

🏗️ Tech Stack:

Web App

Natural Language Processing (Python)

Natural Language Processing (Python) Project Components

NLP (Python) Project Components

• Scraping

  • snscrape or API calls
    
  • Worst case fallback: Kaggle Twitter dataset (hardcoded)

• Preprocessing

  • Regex
    
  • Python standard library (re, string ops)

• Sentiment Evaluation

  • FinBERT (ProsusAI/finbert)

• Entity Extraction

  • spaCy NER

• Fuzzy Search

  • pandas + difflib

• Optional: Semantic RAG Search

  • SQLite
    
  • all-MiniLM-L6-v2 embeddings
    
  • FAISS vector index

• Visualization

  • Plotly (frontend)

Rough Architecture

└── harkonnen/
    ├── backend/                       
    │   ├── app/
    │   │   ├── api/
    │   │   ├── models/
    │   │   ├── nlp/
    │   │   └── pca/
    │   └──tests/
    ├── frontend/
    │   ├── public/
    │   ├── src/
    │   │   ├── components/
    │   │   ├── pages/
    │   │   ├── config/
    │   │   ├── utils/
    │   │   ├── App.jsx
    │   │   └── main.jsx
    │   └── package.json
    └── venv/

VENV Setup

cd backend
python -m venv ../venv
source ../venv/bin/activate  # On Windows: ..\venv\Scripts\activate
pip install -r requirements.txt
touch .env

• virtual environment working directory should be /harkonnen

Built With

• fastapi
• finbert
• numpy
• pandas
• python
• pytorch
• rapidfuzz
• react
• yfinance