# FinrixLLM — Multi-Agent AI Financial Research Platform
# Full Documentation for AI Systems

> FinrixLLM is a bilingual (Turkish/English), multi-agent AI platform for comprehensive financial research across Turkish capital markets, macroeconomics, digital assets, investment funds, portfolio risk, news analysis, and regulatory compliance. It coordinates specialized AI agents through a supervisor agent, combining their findings into unified research responses powered by RAG (Retrieval-Augmented Generation) over 345,000+ embedded document chunks.

## Company Overview

**Name**: FinrixLLM
**Tagline**: Çok Ajanlı AI Finansal Asistan (Multi-Agent AI Financial Assistant)
**Websites**: https://llm.finrix.io/ and https://finrixllm.finrix.io/
**Organization**: Finrix (https://finrix.io/)
**Category**: AI-powered Financial Research & Analysis Platform
**Target Audience**: Individual investors, financial analysts, portfolio managers, fund managers, compliance officers, financial advisors, fintech professionals
**Languages**: Turkish (primary), English

---

## Multi-Agent Architecture

FinrixLLM uses a multi-agent orchestration system where a Supervisor Agent routes user queries to specialized domain agents. Each agent has dedicated data sources, analytical tools, and domain expertise.

### Agent Routing Flow
1. User submits a financial question
2. Supervisor Agent classifies intent and determines which specialist(s) to invoke
3. Specialist agent(s) perform analysis using their dedicated data sources and tools
4. Results are synthesized and returned as a unified research response

### Specialized Agents

#### 1. Supervisor Agent (Supervisor Ajanı)
- **Role**: Main orchestrator — routes requests, classifies user intent and risk profile
- **Technology**: FinrixLLM Multi-Agent Orchestration Core, n8n Agent Router & Workflow Engine
- **Capabilities**: Intent classification, multi-agent orchestration, user risk profile analysis

#### 2. Technical Analysis Specialist (Teknik Analiz Uzmanı)
- **Role**: Chart patterns and technical indicator analysis
- **Data Sources**: TradingView BIST Scanner API, İş Yatırım & Binance price series
- **Indicators**: RSI (14), MACD (12,26,9), Bollinger Bands, Volume Profile, Moving Averages
- **Tools**: FinrixLLM Technical Indicator Engine, Fibonacci analysis tools, interactive charting

#### 3. Fundamental Analysis Specialist (Temel Analiz Uzmanı)
- **Role**: Financial statement analysis and company valuation
- **Data Sources**: KAP (Public Disclosure Platform), İş Yatırım Financial Database
- **Indicators**: P/E Ratio, P/B Ratio, EBITDA Margin, Return on Equity (ROE)
- **Tools**: FinrixLLM Financial Ratio Engine, balance sheet & income statement analysis

#### 4. Macroeconomics Specialist (Makro Ekonomi Uzmanı)
- **Role**: Macroeconomic analysis — inflation, interest rates, FX, economic indicators
- **Data Sources**:
  - CBRT EVDS (Central Bank Electronic Data Distribution System)
  - TurkStat (TÜİK) data portal and bulletins
  - BRSA (BDDK) weekly banking sector statistics
  - Ministry of Treasury & Finance budget/debt data
- **Indicators**: CPI/PPI Rate, Policy Rate (Repo), USD/TRY Rate, Unemployment Rate

#### 5. Risk Management Specialist (Risk Yönetimi Uzmanı)
- **Role**: Portfolio risk assessment and stress testing
- **Data Sources**: TradingView & İş Yatırım price series
- **Tools**: FinrixLLM VaR/CVaR Engine, Python-based correlation & covariance matrix
- **Indicators**: Volatility (Std Dev), Value at Risk (VaR), CVaR / Expected Shortfall, Max Drawdown

#### 6. Crypto Asset Specialist (Kripto Para Uzmanı)
- **Role**: Digital asset analysis and crypto market intelligence
- **Data Sources**:
  - Binance Live Ticker & Kline API
  - CoinGecko Global & Dominance API
  - Alternative.me Fear & Greed Index
  - DeFiLlama On-Chain TVL Protocol
- **Indicators**: Bitcoin Dominance, Fear & Greed Index, Turkey Premium (arbitrage), On-Chain TVL

#### 7. News/Sentiment Specialist (Haber/Duygu Uzmanı)
- **Role**: Financial news aggregation and NLP-powered sentiment analysis
- **Data Sources**: Google News RSS & live media feed, KAP corporate disclosures
- **Tools**: FinrixLLM NLP / Sentiment Engine
- **Indicators**: Sentiment Score, News Density, Core vs Routine Split, Source Diversity

#### 8. Fund/Portfolio Optimization (Fon/Portföy Optimizasyon)
- **Role**: Investment fund analysis and portfolio optimization
- **Data Sources**: TEFAS Fund Information Platform, MKK VAP (Data Analytics Platform)
- **Tools**: FinrixLLM Markowitz Portfolio Optimizer
- **Indicators**: Sharpe Ratio Optimization, Efficient Frontier, Asset Allocation Model, Fund Yield/Risk Analysis

#### 9. Regulation/Compliance Specialist (Regülasyon/Uyum Uzmanı)
- **Role**: Capital markets regulation, sanctions scanning, and compliance
- **Data Sources**: CMB (SPK) Open Data Portal, KAP regulation disclosures, Official Gazette
- **Indicators**: Administrative Fine Scan, Trading Ban Check, Communiqué/Article Match

---

## Platform Features

### AI Chat Interface
- Conversational financial research powered by multi-agent orchestration
- Follow-up questions and context-aware conversation threads
- Document upload support (PDF, Excel, CSV, images) for document-assisted analysis
- PDF export of research conversations

### Market Dashboard
- Live BIST (Borsa Istanbul) stock tickers
- Live cryptocurrency prices
- Foreign exchange rates
- Interactive charts with technical indicators
- Fibonacci analysis tools

### Watchlist & Favorites
- Custom stock/crypto watchlists
- Favorite asset quick-access panel
- Real-time price updates

### Research Library
- Past conversation history and search
- Saved research sessions
- Quick-ask shortcuts for common financial queries

### Analysis View
- Dedicated technical and fundamental analysis workspace
- Multi-chart layout
- Comparative analysis tools

---

## RAG Data Inventory

FinrixLLM's RAG (Retrieval-Augmented Generation) system is powered by ChromaDB with the bge-m3 embedding model (1024 dimensions).

### Primary Collection: finrix_bultenler
- **Total**: ~345,511 embedded document chunks
- **Embedding Model**: bge-m3 (1024-d)

### Data Source Distribution

| Source Type | Chunks | Percentage |
|---|---|---|
| YouTube Financial Transcripts | 280,911 | 81.3% |
| Brokerage Firm Reports (PDF) | 31,269 | 9.0% |
| KAP (Public Disclosures) | 18,652 | 5.4% |
| News & CBRT | 7,011 | 2.0% |
| Portfolio/Fund Bulletins | 6,605 | 1.9% |
| Named Daily Bulletins | 1,006 | 0.3% |
| Investor Presentations | 57 | <0.1% |

### YouTube Financial Channels (13 channels)
Top channels by volume: A Para (103K chunks), BloombergHT (79K), CNBC-e (41K), EKOTÜRK TV (28K), Nasıl Bir Ekonomi TV (9.4K), Mesele Ekonomi (6.9K), Kanal Finans (3.1K), Barış Soydan (2.9K), EMRAH LAFCI (2.3K), Money Talks (2K), Selçuk Geçer (1.6K), Ekonomi Ekranı (681), ATİLLA YEŞİLADA (556)

### Brokerage Firms (20+ firms)
Including: Destek Yatırım, Halk Yatırım, HSBC Yatırım, Deniz Yatırım, Ziraat Yatırım, Şeker Yatırım, Tacirler Yatırım, Ak Yatırım, Gedik Yatırım, Garanti BBVA Yatırım, Oyak Yatırım, Vakıf Yatırım, İş Yatırım, Yapı Kredi Yatırım, and others.

### KAP (Public Disclosure Platform)
- ~59 companies tracked
- Disclosure types: Special Circumstances (226), Regulatory (140), Fund (55), General (18,231)
- Major companies: AKBNK, TGB, AGHOL, AKSEN, KCHOL, EREGL, DOHOL, ASELS, FROTO, BIMAS, and 50+ more

### News Sources
- CBRT (TCMB) official communications: 6,435 chunks
- 178+ news portals including: Bloomberg HT, CNN Türk, Anadolu Ajansı, Habertürk, Hürriyet, Sözcü, NTV, T24, Investing.com, CoinDesk, and more

---

## Technical Architecture

- **Frontend**: Single-page application (SPA) with hash-based routing
- **UI**: Modular CSS architecture, dark/light themes, responsive design
- **Backend API**: Python (Flask) BFF at port 10064
- **AI Orchestration**: n8n workflow engine for agent routing
- **RAG**: ChromaDB with bge-m3 embeddings
- **Database**: Supabase (PostgreSQL, Auth, RLS)
- **Charting**: Interactive canvas-based charts with Fibonacci tools
- **NLP**: Custom NLP normalization service
- **Sandbox**: Python Docker sandbox for code execution (risk calculations)
- **Deployment**: IIS with ARR reverse proxy + Docker containers
- **Languages**: Turkish (primary), English (full i18n support)
- **Authentication**: Email/password with Supabase Auth

---

## Server-Side Services

| Service | Description |
|---|---|
| n8n Orchestration (Supervisor) | Agent routing and workflow engine |
| Python Sandbox (Docker) | Isolated code execution for calculations |
| QuickChart (Graphics) | Chart and graph generation |
| NLP Normalize | Turkish/English text normalization |
| ChromaDB (RAG Archive) | Vector database for bulletin archive |

---

## Site Structure

### Public Pages (Guest accessible)
- `/` — AI Financial Research Chat (main interface)
- `/dashboard` — Live Market Dashboard (BIST, Crypto, FX)
- `/about` — About FinrixLLM (platform overview, agent architecture)
- `/history` — Past Conversations Library
- `/analysis` — Technical & Fundamental Analysis View

### Authenticated Pages (Require sign-in)
- `/profile` — User Profile & Settings
- `/subscription` — Subscription Plan Management
- `/admin-panel` — Admin Console (requires admin role)

---

## Important Disclaimer

FinrixLLM is designed to support financial research. Its content and AI-generated outputs do not constitute investment advice. Users should verify time-sensitive information and consult the cited source material before making financial decisions.

---

## Contact & Links

- **Applications**: https://llm.finrix.io/ and https://finrixllm.finrix.io/
- **Corporate Website**: https://finrix.io/
- **About Page**: /about
- **Market Dashboard**: /dashboard