🤖 Volfefe Machine

Automated Market Volatility Trading System - Detecting and trading on political market signals in real-time

---

🎯 What is Volfefe Machine?

Volfefe Machine is an intelligent, event-driven trading system that monitors real-time political and economic content, analyzes market impact using ML-based sentiment analysis, and executes automated trading strategies based on detected volatility signals.

The name? A playful nod to Trump's infamous "covfefe" tweet + volatility (vol) = Volfefe ⚡

Core Concept

Political/Economic Event → Sentiment Analysis → Market Impact Assessment → Automated Trade

Starting with Truth Social posts (particularly Trump's tariff announcements), the system will expand to monitor news APIs, social media, and financial feeds to identify market-moving events before they cause significant price action.

---

🏗️ Architecture

┌─────────────────────────────────────────────────────────┐
│  SOURCES (Modular Adapters)                             │
│  • Truth Social (via Apify)   • NewsAPI   • RSS   • More│
└─────────────────────────────────────────────────────────┘
                            ↓
┌─────────────────────────────────────────────────────────┐
│  INGESTION PIPELINE                                      │
│  Fetch → Normalize → Store (Postgres) → Broadcast       │
└─────────────────────────────────────────────────────────┘
                            ↓
┌─────────────────────────────────────────────────────────┐
│  MULTI-MODEL CLASSIFICATION (ML Analysis) ✅             │
│  • 3 Sentiment Models: DistilBERT, Twitter-RoBERTa,     │
│    FinBERT (weighted consensus)                          │
│  • 1 NER Model: BERT-base-NER (entity extraction)       │
│  Output: Sentiment + Confidence + Entities (ORG/LOC/PER)│
└─────────────────────────────────────────────────────────┘
                            ↓
┌─────────────────────────────────────────────────────────┐
│  ASSET LINKING (Phase 2 - In Progress)                  │
│  Match entities → Assets database → ContentTargets      │
└─────────────────────────────────────────────────────────┘
                            ↓
┌─────────────────────────────────────────────────────────┐
│  STRATEGY ENGINE (Rule-Based Decisions)                 │
│  Sector Mapping → Company Selection → Trade Type        │
└─────────────────────────────────────────────────────────┘
                            ↓
┌─────────────────────────────────────────────────────────┐
│  EXECUTION (Alpaca API)                                  │
│  Paper Trading → Live Trading (Options, Stocks)         │
└─────────────────────────────────────────────────────────┘

---

🚀 Quick Start

Prerequisites

• Elixir 1.15+ and Erlang/OTP 26+
• PostgreSQL 14+
• Node.js 18+ (for Phoenix LiveView assets)

Installation

# Clone the repository
git clone https://github.com/razrfly/volfefe.git
cd volfefe

# Install dependencies
mix deps.get
cd assets && npm install && cd ..

# Set up environment variables
cp .env.example .env
# Edit .env with your actual credentials (database password, API tokens, etc.)

# Set up database
mix ecto.setup

# (Optional) Install Python dependencies for ML scripts
python3 -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate
pip install -r requirements.txt

# Start Phoenix server
mix phx.server

Visit localhost:4002 to see the live dashboard.

Running Classification with Entity Extraction

Once content is ingested (see Content Ingestion below), you can run multi-model classification:

# Classify first 10 unclassified items with all models (sentiment + NER)
mix classify.contents --limit 10 --multi-model

# Classify all unclassified content
mix classify.contents --all --multi-model

# Classify specific content IDs
mix classify.contents --ids 1,2,3 --multi-model

# Preview what would be classified (dry run)
mix classify.contents --limit 10 --dry-run

Output includes:

• Sentiment consensus from 3 models (positive/negative/neutral)
• Confidence scores and model agreement rates
• Extracted entities: Organizations (ORG), Locations (LOC), Persons (PER), Miscellaneous (MISC)
• Entity confidence scores and context

Market Impact Analysis

After classifying content, analyze market reactions by capturing price/volume snapshots around each Trump post.

1. Initial Setup

First, fetch asset information and establish baseline statistics:

# Fetch starter universe of assets (SPY, QQQ, DIA, IWM, GLD, TLT)
mix fetch.assets --symbols SPY,QQQ,DIA,IWM,GLD,TLT

# Calculate 60-day baseline statistics (mean, std dev, percentiles)
# This fetches historical data and computes rolling returns for 1hr, 4hr, 24hr windows
mix calculate.baselines --all

2. Capture Market Snapshots

Capture 4 time-windowed snapshots (before, 1hr, 4hr, 24hr after) for classified content:

# Single content item
mix snapshot.market --content-id 165

# Multiple specific items
mix snapshot.market --ids 165,166,167

# All content published on a date
mix snapshot.market --date 2025-10-28

# Date range
mix snapshot.market --date-range 2025-10-01 2025-10-31

# All classified content
mix snapshot.market --all

# Only content missing complete snapshots
mix snapshot.market --missing

# Preview without capturing (dry run)
mix snapshot.market --date 2025-10-28 --dry-run

Each snapshot captures:

• Open, High, Low, Close (OHLC) prices
• Volume and volume deviation from baseline
• Volume z-score (standard deviations from mean)
• Market state (pre-market, regular, after-hours, closed)
• Data validity flags
• Isolation score (contamination from nearby content)

3. Update Baselines

Keep baseline statistics fresh as new market data accumulates:

# Update only stale baselines (older than 24 hours)
mix calculate.baselines --all --check-freshness

# Update specific assets
mix calculate.baselines --symbols SPY,QQQ

# Force recalculation (ignore freshness)
mix calculate.baselines --all --force

4. Data Validation

Monitor data quality and coverage:

# Check which content is missing snapshots
mix snapshot.market --missing --dry-run

# View baseline statistics
iex -S mix
> alias VolfefeMachine.{Repo, MarketData.BaselineStats}
> Repo.all(BaselineStats) |> Enum.map(& {&1.asset_id, &1.window_minutes, &1.mean_return})

Common Workflows

Daily Update Workflow:

# 1. Update baselines for all assets (skips fresh ones)
mix calculate.baselines --all --check-freshness

# 2. Capture snapshots for newly classified content
mix snapshot.market --missing

# 3. Verify coverage
mix snapshot.market --missing --dry-run

Backfill Historical Data:

# Capture snapshots for all content in October 2025
mix snapshot.market --date-range 2025-10-01 2025-10-31

# Force recalculate all baselines with latest 60 days
mix calculate.baselines --all --force

Troubleshooting:

• "No data available": TwelveData may not have data for that timestamp (market closed, weekend, holiday)
• "Rate limit exceeded": Free tier allows 8 calls/minute, 800/day - add delays between operations
• Incomplete snapshots: Use --missing flag to find and fill gaps
• Stale baselines: Use --check-freshness to update only old statistics

Environment Variables

The project uses environment variables for sensitive configuration. Copy .env.example to .env and update with your credentials:

# PostgreSQL Database
PGHOST=localhost
PGDATABASE=volfefe_machine_dev
PGUSER=postgres
PGPASSWORD=your_postgres_password

# Apify API (for Truth Social scraping)
APIFY_USER_ID=your_user_id_here
APIFY_PERSONAL_API_TOKEN=your_api_token_here

⚠️ Never commit your .env file to version control! It's already in .gitignore.

Content Ingestion

Status: ✅ Complete - Unified ingestion pipeline ready

Fetch and import content from Truth Social using a single command:

# Fetch 100 posts from a specific user
mix ingest.content --source truth_social --username realDonaldTrump --limit 100

# Include replies in results
mix ingest.content --source truth_social --username realDonaldTrump --limit 50 --include-replies

# Preview what would be fetched (dry run)
mix ingest.content --source truth_social --username realDonaldTrump --limit 10 --dry-run

Available Options:

• --source, -s - Content source (currently: truth_social)
• --username, -u - Username/profile to fetch (required)
• --limit, -l - Maximum posts to fetch (default: 100)
• --include-replies - Include replies in results (default: false)
• --dry-run - Preview configuration without fetching (default: false)

---

🧩 Key Components

Component	Purpose	Status
Database Schema	Assets, Contents, Classifications, ContentTargets	✅ Complete
Multi-Model Classification	3 sentiment models + weighted consensus	✅ Complete
NER Entity Extraction	Extract organizations, locations, persons	✅ Complete
Apify Integration	Fetch Truth Social posts via API	✅ Complete
Ingestion Pipeline	Unified fetch + import workflow	✅ Complete
Asset Linking	Match extracted entities to assets database	📋 Phase 2
Strategy Engine	Rule-based trade decision logic	📋 Phase 3
Trade Executor	Alpaca API integration	📋 Phase 4
Dashboard	Real-time monitoring UI	📋 Future

Legend: ✅ Complete | 🚧 In Progress | 📋 Planned

---

🗄️ Data Model

Core Tables

sources - External data sources (Truth Social, NewsAPI, etc.)

%Source{
  name: "truth_social",
  adapter: "TruthSocialAdapter",
  base_url: "https://api.example.com",
  last_fetched_at: ~U[2025-01-26 10:00:00Z]
}

contents - Normalized posts/articles

%Content{
  source_id: uuid,
  external_id: "12345",
  author: "realDonaldTrump",
  text: "Big tariffs on steel coming soon!",
  url: "https://truthsocial.com/@realDonaldTrump/12345",
  published_at: ~U[2025-01-26 09:45:00Z],
  classified: false
}

classifications - ML analysis results with sentiment consensus

%Classification{
  content_id: uuid,
  sentiment: "negative",
  confidence: 0.9556,
  meta: %{
    "agreement_rate" => 1.0,
    "model_results" => [
      %{"model_id" => "distilbert", "sentiment" => "negative", "confidence" => 0.9812},
      %{"model_id" => "twitter_roberta", "sentiment" => "negative", "confidence" => 0.9654},
      %{"model_id" => "finbert", "sentiment" => "negative", "confidence" => 0.9201}
    ],
    "entities" => [
      %{"text" => "Tesla", "type" => "ORG", "confidence" => 0.9531},
      %{"text" => "United States", "type" => "LOC", "confidence" => 0.9912}
    ]
  }
}

assets - Tradable securities (9,000+ loaded)

%Asset{
  symbol: "TSLA",
  name: "Tesla Inc",
  exchange: "NASDAQ",
  asset_class: "us_equity"
}

content_targets - Extracted entities linked to assets (Phase 2)

%ContentTarget{
  content_id: uuid,
  asset_id: uuid,
  extraction_method: "ner_bert",
  confidence: 0.9531,
  context: "Tesla stock tumbled 12% today..."
}

---

🛠️ Tech Stack

Backend

• Framework: Phoenix 1.7 + LiveView
• Language: Elixir
• Database: PostgreSQL with Ecto
• Job Queue: Oban for background processing
• HTTP Client: HTTPoison for external APIs

Machine Learning

• Python: Python 3.9+ with virtual environment
• ML Framework: Transformers (Hugging Face)
• Models:
  • Sentiment: DistilBERT, Twitter-RoBERTa, FinBERT
  • NER: BERT-base-NER (dslim/bert-base-NER)
• Elixir Integration: Python interop via System.cmd/3

External Services

• Data Source: Apify for Truth Social scraping
• Trading: Alpaca Markets API (future)

---

📅 Roadmap

Phase 1: Foundation & ML Pipeline ✅ (Complete)

• Project setup and architecture
• Database schemas (contents, sources, classifications, assets, content_targets)
• Assets database loaded (9,000+ securities)
• Multi-model sentiment classification (DistilBERT, Twitter-RoBERTa, FinBERT)
• Weighted consensus algorithm
• NER entity extraction (BERT-base-NER)
• Classification mix task with batch processing
• Content ingestion - Unified mix task (Issue #46)
• Content backup/seeding system (Issue #45) - Next Step

Phase 2: Asset Linking (In Progress)

• Entity → Asset matching logic (Issue #42)
• ContentTargets creation
• Fuzzy name matching
• Confidence scoring
• Manual validation tools

Phase 3: Strategy Engine

• Sector-to-ticker mapping
• Rule-based trade logic
• Backtesting framework
• Signal generation

Phase 4: Trade Execution

• Alpaca API integration
• Paper trading
• Risk management
• Live trading (manual approval)

Phase 5: Multi-Source Expansion

• NewsAPI adapter
• Reddit adapter
• RSS feeds
• Source weighting

See: Issue #43 (Phase 1 NER) | Issue #42 (Phase 2 Asset Linking)

---

🎲 Polymarket Insider Detection

The system includes tools for detecting potential insider trading on Polymarket prediction markets by analyzing blockchain trade data.

Architecture

Reference Cases (Seed Data)      Blockchain (Subgraph)
├─ Event date                    ├─ ALL trades in date range
├─ Description                   ├─ Wallet addresses
└─ Context from news             └─ Trade amounts & timing
        │                                   │
        └───────────┬───────────────────────┘
                    ▼
            Pattern Discovery
            ├─ Scan by date range (not keyword)
            ├─ Group trades by market
            ├─ Score: volume, whales, timing
            └─ Output candidate markets
                    ▼
            Human Confirmation
            ├─ Review candidate list
            ├─ Confirm correct market
            └─ Update reference case
                    ▼
            Trade Ingestion
            └─ Ingest trades for analysis

Key Commands

1. Scan Mode (Discovery Without Reference Case)

Scan blockchain trades in a date range to discover market activity patterns:

# Scan last 7 days, show top 10 markets
mix polymarket.ingest --subgraph --days 7 --scan

# Scan specific date range around a known event
mix polymarket.ingest --subgraph --from 2025-10-08 --to 2025-10-12 --scan --top 20

# Full ingestion (without --scan flag)
mix polymarket.ingest --subgraph --from 2025-10-01 --to 2025-10-15

Scan output includes:

• Total trades and volume in date range
• Markets grouped by trading activity
• Whale trade counts (>$1K positions)
• Unique wallet counts per market
• Trading period timestamps

2. Reference Case Discovery (Find Market for Known Event)

Discover which Polymarket market corresponds to a reference case:

# Discover markets for a specific reference case
mix polymarket.discover --reference-case "Nobel Peace Prize 2025"

# Custom window: 10 days before event, 2 days after
mix polymarket.discover --reference-case "Nobel Peace Prize 2025" --window 10 --after 2

# Show top 20 candidates
mix polymarket.discover --reference-case "Nobel Peace Prize 2025" --top 20

# Discover for ALL reference cases missing condition_ids
mix polymarket.discover --all-references

Discovery output includes:

• Ranked candidate markets by score (0.0-1.0)
• Condition ID for each candidate
• Volume and pre-event volume percentage
• Whale count and unique wallet count
• Trading period around the event
• Suspicious wallets with timing and volume data (Phase 3)

3. Confirm Market Match

After reviewing discover output, confirm which market is correct:

# Confirm market match for reference case
mix polymarket.confirm --reference-case "Nobel Peace Prize 2025" --condition 0x14a3dfeba8...

# With optional slug (auto-fetched if omitted)
mix polymarket.confirm --reference-case "Case Name" --condition 0xabc... --slug "market-slug"

4. Ingest Trades

After confirming, ingest trades for analysis:

# Ingest trades for a specific reference case
mix polymarket.ingest --subgraph --reference-case "Nobel Peace Prize 2025"

# Ingest trades for all reference cases with condition_ids
mix polymarket.ingest --subgraph --reference-cases

# Ingest for specific condition
mix polymarket.ingest --subgraph --condition 0x14a3dfeba8... --from 2025-10-01

Complete Workflow Example

# 1. You have a reference case (from news: "Nobel Peace Prize 2025", event Oct 11)
#    Check it exists in the database
mix polymarket.references

# 2. Discover which market matches this event
mix polymarket.discover --reference-case "Nobel Peace Prize 2025"
# Output shows candidate markets ranked by score:
#   1. "Will María Corina Machado win the Nobel Peace Prize?"
#      Condition: 0x14a3...  Score: 0.85  Volume: $45K (62% pre-event)

# 3. Review candidates and confirm the correct match
mix polymarket.confirm --reference-case "Nobel Peace Prize 2025" --condition 0x14a3dfeba8...

# 4. Promote discovered wallets to investigation candidates
mix polymarket.promote --reference-case "Nobel Peace Prize 2025"
# Output: Created 12 investigation candidate(s)
#         Batch ID: refcase-nobel-peace-1234567890

# 5. Ingest trades for detailed analysis
mix polymarket.ingest --subgraph --reference-case "Nobel Peace Prize 2025"

# 6. Review and investigate candidates
mix polymarket.candidates --batch refcase-nobel-peace-1234567890
mix polymarket.investigate --id 1

Scoring Algorithms

Market Scoring

Reference case discovery scores markets based on:

Factor	Weight	Description
Whale Activity	30%	Trades >$1K indicate informed trading
Pre-Event Volume	30%	Volume concentration before event
Total Volume	25%	Log-scaled trading volume
Wallet Diversity	15%	Unique wallets (avoid wash trading)

Wallet Suspicion Scoring (Phase 3)

Individual wallets are scored for suspicious activity:

Factor	Weight	Description
Volume	25%	Log-scaled position size
Whale Trades	25%	Number of trades >$1K
Pre-Event Concentration	30%	% of volume placed before event
Timing Precision	20%	Closer to event = more suspicious

Timing Precision Breakdown:

• Within 24h of event: 20%
• Within 48h: 15%
• Within 72h: 10%
• Beyond 72h: 5%

Discovery results (wallets + candidate markets) are automatically saved to the reference case for later analysis.

5. Promote Wallets to Investigation Candidates (Phase 4)

After discovery, convert suspicious wallets into investigation candidates:

# Preview promotion (dry run)
mix polymarket.promote --reference-case "Nobel Peace Prize 2025" --dry-run

# Promote wallets (creates investigation candidates)
mix polymarket.promote --reference-case "Nobel Peace Prize 2025"

# Custom thresholds
mix polymarket.promote --reference-case "Case Name" --min-score 0.6 --limit 10

# Force priority level
mix polymarket.promote --reference-case "Case Name" --priority critical

Promotion workflow:

1. Reads discovered wallets from reference case
2. Filters by minimum suspicion score (default: 0.4)
3. Creates InvestigationCandidate records with:
   • Anomaly breakdown (volume, timing, whale trades)
   • Matched patterns (reference case linkage)
   • Priority based on suspicion score
4. Assigns batch ID for tracking

After promotion:

# List candidates from this batch
mix polymarket.candidates --batch refcase-nobel-peace-1234567890

# Investigate specific candidate
mix polymarket.investigate --id 1

# Confirm as insider
mix polymarket.confirm --id 1 --notes "Pre-event timing matches"

Data Sources

Source	Endpoint	Data Available
Polymarket Subgraph	api.goldsky.com/.../orderbook-subgraph	All trades since Nov 2022
Polymarket API	data-api.polymarket.com	Recent trades (geo-blocked in US)
Gamma API	gamma-api.polymarket.com	Market metadata (geo-blocked in US)
CLOB API	clob.polymarket.com	Active market data (geo-blocked in US)

Note: The subgraph bypasses geo-blocking and provides complete historical data. For CLOB and Gamma APIs, see VPN Setup below.

VPN Setup (Required for US Users)

Polymarket geo-blocks US IP addresses for their CLOB and Gamma APIs (regulatory compliance). The application uses a Docker-based VPN proxy to route only Polymarket API calls through VPN, without affecting the rest of your network traffic.

Prerequisites

• Docker Desktop installed and running
• ProtonVPN account with WireGuard access

Setup Steps

1. Get WireGuard credentials from ProtonVPN:

   • Go to Proton Account → VPN → WireGuard
   • Click "Generate Key" (or use existing)
   • Copy the PrivateKey value

2. Add credentials to .env:

   # VPN Proxy for Polymarket API Access
   PROTONVPN_WIREGUARD_PRIVATE_KEY=your_wireguard_private_key_here
   VPN_PROXY_ENABLED=true
   VPN_PROXY_HOST=localhost
   VPN_PROXY_PORT=8888

3. Start the VPN proxy container:

   docker compose -f docker-compose.vpn.yml up -d

4. Verify connection:

   # Check container is running
   docker logs vpn-proxy | grep "Public IP"
   # Should show: Public IP address is X.X.X.X (Netherlands, ...)
   
   # Test API access through proxy
   curl -x http://localhost:8888 "https://gamma-api.polymarket.com/markets?limit=1"

Usage

When the VPN proxy is running and VPN_PROXY_ENABLED=true, the application automatically routes Polymarket API calls (CLOB, Gamma) through the VPN tunnel. The subgraph API does not require VPN.

# Start VPN proxy
docker compose -f docker-compose.vpn.yml up -d

# Run with VPN enabled
export VPN_PROXY_ENABLED=true
mix phx.server

# Or for one-off commands
export VPN_PROXY_ENABLED=true
mix polymarket.health

Troubleshooting

Issue	Solution
Container won't start	Ensure Docker Desktop is running
"Connection refused" on port 8888	Check docker logs vpn-proxy for errors
API still returning 403/blocked	Verify VPN connected: docker logs vpn-proxy | grep "Public IP"
Rate limiting from Polymarket	The VPN is working; reduce request frequency
Enrichment/metadata fetch failing	Ensure VPN_PROXY_ENABLED=true is set

Stopping the VPN

docker compose -f docker-compose.vpn.yml down

Note: When the VPN proxy is stopped or VPN_PROXY_ENABLED=false, CLOB and Gamma API calls will fail for US users. The subgraph-based trade ingestion will continue to work.

---

🧪 Testing

# Run all tests
mix test

# Run with coverage
mix test --cover

# Run specific test file
mix test test/volfefe/pipeline_test.exs

---

📊 Entity Extraction Output Example

Input Text:

"Tesla stock tumbled 12% today as Elon Musk's controversial tweet sparked
concerns about the company's future. Analysts in the United States and
Europe are worried about automotive sector stability."

Multi-Model Classification Output:

%{
  # Sentiment Consensus (3 models)
  consensus: %{
    sentiment: "negative",
    confidence: 0.9556,
    agreement_rate: 1.0
  },

  # Individual Model Results
  model_results: [
    %{model_id: "distilbert", sentiment: "negative", confidence: 0.9812},
    %{model_id: "twitter_roberta", sentiment: "negative", confidence: 0.9654},
    %{model_id: "finbert", sentiment: "negative", confidence: 0.9201}
  ],

  # Extracted Entities (NER)
  entities: [
    %{text: "Tesla", type: "ORG", confidence: 0.9531,
      context: "Tesla stock tumbled 12% today..."},
    %{text: "Elon Musk", type: "PER", confidence: 0.9802,
      context: "...12% today as Elon Musk's controversial..."},
    %{text: "United States", type: "LOC", confidence: 0.9912,
      context: "...Analysts in the United States and Europe..."},
    %{text: "Europe", type: "LOC", confidence: 0.9845,
      context: "...United States and Europe are worried..."}
  ],

  # Entity Statistics
  entity_stats: %{
    total_entities: 4,
    by_type: %{"ORG" => 1, "LOC" => 2, "PER" => 1, "MISC" => 0}
  },

  # Performance
  total_latency_ms: 663,
  successful_models: 4
}

Phase 2 Preview (not yet implemented):

• "Tesla" → Match to Asset{symbol: "TSLA", name: "Tesla Inc"}
• Create ContentTarget{content_id: X, asset_id: Y, confidence: 0.95}

---

📊 Example Pipeline Flow

Current Workflow (Manual)

1. Fetch & Import - mix ingest.content --source truth_social --username USER --limit 100
2. Content Storage - Posts stored in PostgreSQL contents table
3. Multi-Model Classification - Run mix classify.contents --all --multi-model
   • 3 sentiment models analyze text (DistilBERT, Twitter-RoBERTa, FinBERT)
   • Weighted consensus calculates final sentiment + confidence
   • NER model extracts entities (ORG, LOC, PER, MISC)
4. Results Storage - Classifications saved to classifications table
5. Entity Analysis - Entities stored in classification metadata

Future Automated Workflow

1. Scheduler (Oban) - Poll Truth Social every 60 seconds
2. Adapter - Fetch and normalize new posts
3. PubSub - Broadcast {:new_content, content} events
4. Auto-Classification - Trigger multi-model analysis on new content
5. Asset Linking (Phase 2) - Match entities to assets, create ContentTargets
6. Strategy Engine (Phase 3) - Generate trade recommendations
7. Executor (Phase 4) - Place orders via Alpaca API
8. Dashboard - Real-time monitoring via LiveView

---

🤝 Contributing

This is currently a private project, but contributions are welcome! Please:

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'Add amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

---

⚖️ Legal & Risk Disclaimer

This software is for educational and research purposes only.

• Automated trading carries significant financial risk
• Past performance does not guarantee future results
• This system is not financial advice
• Use at your own risk
• Always start with paper trading
• Understand all risks before deploying real capital

By using this software, you acknowledge that you are solely responsible for any trading decisions and outcomes.

---

📄 License

MIT License - see LICENSE for details

---

🔗 Resources

Framework & Platform

• Phoenix Framework Docs
• Elixir Getting Started
• PostgreSQL Documentation

Machine Learning Models

• Hugging Face Transformers
• DistilBERT (SST-2)
• Twitter-RoBERTa Sentiment
• FinBERT
• BERT-base-NER

APIs & Services

• Apify Documentation
• Alpaca API Documentation

GitHub Issues

• Issue #43: Phase 1 NER Entity Extraction
• Issue #42: Phase 2 Asset Linking
• Issue #45: Content Data Seeding

---

Built with ❤️ and Elixir