Mini Finance Dashboard

Production-oriented macroeconomic and financial dashboard built with Python and Dash, enhanced with a multi-agent AI research system for daily market context and structured news summaries.

This repository is maintained as a non-commercial personal project for research, learning, and product experimentation.

Legal Notice This repository and the in-app Research Hub are designed for research and education. They do not provide licensed investment advice, portfolio management, or user-specific suitability analysis. Outputs are based on public data, public news flow summaries, and AI-generated research text and should not be used as a sole basis for investment decisions. If any AI-generated output drifts into advice-like wording, that wording remains non-advisory and no responsibility is accepted for reliance, trading decisions, or losses. Jurisdiction-specific legal notices in English, Turkish, German, French, Italian, Polish, and Hungarian, plus EU / Switzerland / USA / UK notes, are available in LEGAL_NOTICE.md and on the /terms page.

Disclaimer Order

English Disclaimer

This repository, dashboard, and Research Hub are provided for research and education only. They do not provide licensed investment advice, portfolio management, brokerage, or user-specific suitability analysis. No user's financial situation, risk tolerance, or investment objectives are taken into account. Outputs include AI-generated research text, and if any model output drifts into advice-like wording, it remains non-advisory and no responsibility is accepted for reliance, trading decisions, or losses.

Türkçe Disclaimer - Türkiye

Bu repo, dashboard ve Research Hub yalnızca araştırma ve eğitim amacıyla sunulur. Hiçbir içerik 6362 sayılı Sermaye Piyasası Kanunu kapsamında lisanslı yatırım danışmanlığı, portföy yönetimi veya kişiye özel yatırım tavsiyesi niteliğinde değildir. Kullanıcının finansal durumu, risk toleransı ve yatırım hedefleri dikkate alınmaz. İçeriklerin bir kısmı yapay zeka tarafından üretilen araştırma metinleridir; yapay zeka çıktısı tavsiye benzeri bir ifadeye kaysa bile bu ifade yatırım tavsiyesi sayılmaz ve buna dayanılarak alınan kararlar, işlemler veya zararlar bakımından sorumluluk kabul edilmez.

Resmi Türkiye referansları:

• SPK, Yatırım Hizmet ve Faaliyetleri ile Yatırım Kuruluşlarına İlişkin Rehber: <https://spk.gov.tr/duyurular/basin-duyurulari/2014/yatirim-hizmet-ve-faaliyetleri-ile-yatirim-kuruluslarina-iliskin-rehber>
• SPK, Bilgi Bazlı Piyasa Dolandırıcılığı: <https://spk.gov.tr/yatirimcilar/sermaye-piyasasinda-piyasa-dolandiriciligi-manipulasyon/bilgi-bazli-piyasa-dolandiriciligi>

Deutscher Disclaimer - Deutschland

Dieses Repository, Dashboard und der Research Hub dienen ausschließlich Informations-, Forschungs- und Bildungszwecken. Die Inhalte sind nicht als Anlageberatung im Sinne von § 2 Abs. 2 Nr. 4 WpIG ausgestaltet; persönliche Umstände, Risikotoleranz und Anlageziele werden nicht berücksichtigt. Öffentlich verbreitete anlagebezogene Einschätzungen können zudem an die Transparenz- und Objektivitätsanforderungen der §§ 85-86 WpHG sowie der einschlägigen EU-Regeln anknüpfen. Teile der Inhalte werden KI-generiert erstellt; selbst wenn eine KI-Ausgabe in beratungsähnliche Formulierungen abrutscht, bleibt sie nicht-beratend, und für Vertrauen auf solche Inhalte, Handelsentscheidungen oder Verluste wird keine Verantwortung übernommen.

Praktische Kurzfassung:

"Diese Plattform ist kein zugelassener Anlageberatungsdienst. Alle Inhalte dienen ausschließlich Informations-, Forschungs- und Bildungszwecken."

Offizielle Deutschland-Referenzen:

• WpIG § 2: <https://www.gesetze-im-internet.de/wpig/__2.html>
• WpHG §§ 85-86: <https://www.gesetze-im-internet.de/wphg/__85.html>
• BaFin, Wertpapier-Informationsblatt: <https://www.bafin.de/DE/Aufsicht/Prospekte/Wertpapiere/Wertpapier-Informationsblaetter/wib_node.html>

Other Language Notices

• Français: Cette plateforme n'est pas un service agréé de conseil en investissement. Certains contenus sont générés par IA; même si une formulation ressemble à un conseil, elle reste non consultative et aucune responsabilité n'est assumée. Voir LEGAL_NOTICE.md et /terms pour la version détaillée et les références AMF.
• Italiano: Questa piattaforma non costituisce un servizio autorizzato di consulenza in materia di investimenti. Alcuni contenuti sono generati dall'IA; anche se una formulazione appare simile a un consiglio, resta non consulenziale e non si assume alcuna responsabilità. Vedi LEGAL_NOTICE.md e /terms per la versione dettagliata e i riferimenti CONSOB.
• Polski: Ta platforma nie stanowi licencjonowanej usługi doradztwa inwestycyjnego. Część treści jest generowana przez AI; nawet jeśli jakieś sformułowanie upodobni się do porady, pozostaje ono niewiążące i nie przyjmuje się odpowiedzialności. Zobacz LEGAL_NOTICE.md i /terms, aby uzyskać wersję szczegółową oraz odniesienia KNF.
• Magyar: Ez a platform nem minősül engedéllyel rendelkező befektetési tanácsadási szolgáltatásnak. Egyes tartalmak MI által generáltak; akkor sem minősülnek tanácsadásnak, ha megfogalmazásuk annak tűnik, és felelősséget ezért nem vállalunk. A részletes változatot és az MNB-hivatkozásokat lásd a LEGAL_NOTICE.md fájlban és a /terms oldalon.

Screenshots

Main Dashboard

The main page combines the legal notice banner, country macro KPIs, market board, special date-range controls, and ETF performance blocks in a single Dash interface.

Agentic AI Research Hub

The Research Hub shows a non-advisory sector-level AI summary, daily research snapshot, and Chronos memory context generated by the multi-agent pipeline.

Features

• Compare macro indicators across 10 supported countries.
• Track market assets and sector ETFs with daily cached data.
• Simulate historical investments in FX, gold, indices, ETFs, and interest-bearing products.
• Research Hub — 20 AI agents gather real-time news, produce structured research outlooks, and generate a sector research overview.
• Modular architecture easy to extend with additional data providers or storage backends.

Tech Stack

• Python 3.10+
• Dash for the web app shell, callbacks, routing, and interactive controls
• Plotly for charts and financial/macro visualizations
• Pandas for tabular and time-series transformations
• Requests for World Bank, OECD, ECB, and other HTTP fetchers
• yfinance for market prices, ETFs, FX, and index snapshots
• OpenAI API with web search for the real-time news-agent layer
• Optional Ollama support for non-search research, historian, overview, and translation runs
• PyYAML for ai_config.yaml loading
• SQLite caching for repeatable refreshes and lower external API load

Charts in this project are rendered with Plotly inside Dash. There is no separate Chart.js layer in the current codebase.

Pages

Agentic Research System

The dashboard includes a three-phase multi-agent pipeline. News agents browse the internet for current developments, research agents convert those conclusions into structured sector research summaries, and an overview agent compresses the sector calls into a top-of-page research snapshot.

Architecture

┌─────────────────────────────────────────────────────────────┐
│                     PHASE 1: NEWS AGENTS                    │
│                  (Web Search — Real-Time)                   │
│                                                             │
│  Atlas  Machiavelli  Cicero  Hippocrates  Enzo              │
│  Aegis  Turing  Silicon  Pixel  Ada                         │
└──────────────────────────┬──────────────────────────────────┘
                           │ conclusions passed as context
                           ▼
┌─────────────────────────────────────────────────────────────┐
│                  PHASE 2: RESEARCH AGENTS                  │
│                  (No Web Search — Analysis)                 │
│                                                             │
│  Spartan  Quantum  Nexus  Torque  Muse                      │
│  Vitalis  Volt  Oracle  Keynes                              │
└──────────────────────────┬──────────────────────────────────┘
                           │
                           ▼
┌─────────────────────────────────────────────────────────────┐
│                   PHASE 3: OVERVIEW AGENT                   │
│           (Sector Research Overview — No Search)            │
│                                                             │
│  Beacon                                                     │
└──────────────────────────┬──────────────────────────────────┘
                           │
                           ▼
            Dashboard — Research Overview + Final Reports

News agents run in parallel (2 workers) with automatic retry on rate limits. Once all complete, their combined output is fed as a single context block to the research agents, which also run in parallel. After the research phase finishes, the overview agent creates the snapshot shown at the top of the Research Hub page.

News Agents

These agents use the OpenAI web search API to fetch real-time news from the internet before generating their reports.

Research Agents

These agents receive the combined news conclusions as context and produce research outlooks for their respective domains. They do not access the internet directly.

Overview Agent

Legal and Usage Guardrails

• The Research Hub is designed as a research and education workflow, not a licensed investment advisory product.
• The / dashboard and /agents Research Hub now show persistent legal notices and require session-level acknowledgement before access is granted.
• Prompts and UI have been rewritten to avoid direct buy / hold / sell language and to favor neutral research terminology.
• AI-generated outputs remain non-advisory even if a model drifts into advice-like wording, and no responsibility is accepted for reliance, trades, or losses arising from such output.
• Full usage terms live in TERMS_OF_USE.md.
• Jurisdiction-specific notices and regulatory references live in LEGAL_NOTICE.md.
• The /terms page now exposes both the usage terms and the jurisdiction-specific legal notice.

AI Configuration

Create or edit ai_config.yaml in the project root (already gitignored):

openai:
  api_key: "YOUR_OPENAI_API_KEY_HERE"
  model: "gpt-4o-mini"
  temperature: 0.7
  timeout: 60
  max_tokens_news: 2000
  max_tokens_analyst: 3000

analysis_llm:
  provider: "ollama"   # "openai" or "ollama"
  model: "gemma4:e4b"  # optional; defaults to openai.model when provider=openai
  base_url: "http://localhost:11434/v1"
  api_key: "ollama"

Note: ai_config.yaml is listed in .gitignore and will never be committed. Phase 1 runtime split: news agents still use OpenAI web search. Research, historian, overview, and translation steps can optionally run on local Ollama via analysis_llm.

Available Models & Compatibility

The News / Search Model selector controls the exact OpenAI Chat Completions search runtime used by news agents. The Research / Translation LLM selector controls research agents, the historian, the overview agent, and English translation.

OpenAI Models

Local Ollama Option

For the non-search pipeline, the dashboard can also use:

News Agent Models (OpenAI Chat Completions)

The dashboard currently exposes only the OpenAI Chat Completions search models that officially support web_search_options:

Rate limits: gpt-5-search-api has a 6,000 TPM / 100 RPM limit. The system runs 2 news agents in parallel and retries with exponential backoff if rate-limited. A full run of all 20 agents typically takes 3–8 minutes depending on the model and rate limits.

Supported Countries

Project Structure

finance_dashboard/
  agents/
    prompts/
      news/           ← news agent character prompts (Turkish)
      analysts/       ← research agent prompts (Turkish, kept in same folder for compatibility)
      summary/        ← overview agent prompt (Turkish)
    results/          ← daily run outputs as JSON (gitignored)
    config.py         ← ai_config.yaml loader
    engine.py         ← three-phase agent orchestration
  dashboard/
    app.py            ← main Dash app with URL routing
    agent_page.py     ← page 2 layout and callbacks
  data/               ← World Bank, OECD, ECB, Yahoo Finance fetchers
  domain/             ← data models
  services/           ← macro, market, ETF, simulation services
  simulation/         ← investment simulator
  storage/            ← SQLite cache
  config.py
  country_registry.py
  main.py
assets/
  styles.css
examples/
tests/
ai_config.yaml        ← API key & model config (gitignored)
pyproject.toml

Installation

python -m venv .venv
source .venv/bin/activate
pip install -e .

Run

source .venv/bin/activate
python -m finance_dashboard.main

Open http://127.0.0.1:8050 in your browser.

Click To Run (macOS)

You can also double-click start_dashboard.command from Finder.

The launcher:

• uses the project's .venv
• starts the Dash server
• opens http://127.0.0.1:8050 automatically in your browser

Data Sources

Project context: this repository is maintained as a non-commercial personal project, not as a commercial product or paid data service.

Testing

pytest