The Financial Chronical

Your AI-Powered Investment War Room - Transform Global News into Actionable Investment Intelligence

Comment

💡 Inspiration The inspiration came from a simple observation: information overload is killing investment decisions.

Every day, thousands of news articles are published about companies, markets, and industries. Professional investors have Bloomberg terminals and analyst teams. But what about individual investors? They're drowning in noise, unable to separate signal from chaos.

We asked ourselves: What if AI could be your personal investment analyst?

That's when we discovered Google's Gemini 3 Flash - a model powerful enough to understand complex financial narratives, fast enough to process news in real-time, and intelligent enough to connect the dots across seemingly unrelated stories.

🎓 What We Learned AI is a Reasoning Engine, Not Just a Chatbot

Gemini 3 Flash can perform multi-step reasoning across complex financial scenarios It understands context, sentiment, and competitive dynamics The key is prompt engineering - we built specialized prompts for each analysis layer Knowledge Graphs are Game-Changers

Connecting entities (companies, people, events) reveals hidden patterns Stories evolve over time - tracking maturity (DEVELOPING → MATURE → ACTIONABLE) is crucial Graph-based analysis beats linear news feeds Real-Time is Hard

Web scraping is fragile - sites change, rate limits exist, content varies We built a dynamic scraper that adapts to different news sources Caching and async processing are essential for performance UX Matters in FinTech

Financial data is intimidating - we designed a "newspaper" aesthetic to make it approachable Color-coding (green for bullish, red for bearish) provides instant insights Progressive disclosure - show summaries first, details on demand Gemini 3 Flash is Production-Ready

Fast response times (< 2 seconds for complex analysis) Consistent output quality with structured prompts Cost-effective for real-time applications 🏗️ How We Built It Phase 1: Intelligence Layer (Days 1-2) Built web scrapers for NewsAPI, TechCrunch, Bloomberg Implemented entity extraction using Gemini 3 Flash Created knowledge graph to track companies, people, and events Developed story maturity classification (DEVELOPING → MATURE → ACTIONABLE) Phase 2: Analysis Engine (Days 3-4) Designed multi-layer AI analysis: Cognitive Layer: Investment thesis, conviction scoring, winners/losers Competitive Intelligence: Market positioning, competitive advantages Macro Context: Economic trends, market timing signals Sentiment Analysis: Trend detection across time Integrated Gemini 3 Flash for all AI operations Built comprehensive subreport generation Phase 3: Portfolio War Room (Days 5-6) Created AI-driven portfolio management Implemented signal generation (BUY/HOLD/EXIT/WATCH) Built risk assessment and exit strategy planning Added real-time alerts and opportunity discovery Designed "Investment War Room" UI with story integration Phase 4: Frontend Excellence (Days 7-8) Built modern React 19 + TypeScript frontend Designed newspaper-inspired UI for approachability Created interactive dashboards with Recharts Implemented responsive design for mobile/desktop Added real-time data synchronization Phase 5: Deployment & Polish (Day 9) Deployed backend to Render with auto-scaling Deployed frontend to Vercel with automatic GitHub deployments Configured CORS and environment variables Optimized build process and fixed TypeScript errors Created comprehensive documentation 🚧 Challenges We Faced Challenge 1: Gemini API Rate Limits Problem: Initial implementation hit rate limits during bulk analysis.

Solution:

Implemented request queuing and exponential backoff Added caching layer to avoid redundant API calls Batched similar requests together Challenge 2: Inconsistent Web Scraping Problem: Different news sites have different HTML structures.

Solution:

Built a dynamic scraper that adapts to different sources Used Newspaper3k for article extraction with fallbacks Implemented robust error handling and retry logic Challenge 3: Knowledge Graph Complexity Problem: Tracking relationships between entities across time is complex.

Solution:

Designed a story-centric graph structure Each story tracks its own entities and evolution Used NetworkX for graph operations and analysis Challenge 4: Real-Time Portfolio Updates Problem: Portfolio needs to reflect current market conditions and AI signals.

Solution:

Created enhanced portfolio endpoints with AI integration Linked portfolio positions to driving stories Implemented real-time signal calculation based on sentiment trends Challenge 5: TypeScript Build Errors in Production Problem: Vercel deployment failed due to unused imports.

Solution:

Set up local build testing before deployment Fixed TypeScript strict mode errors Configured proper build pipeline with type checking Challenge 6: UX for Complex Financial Data Problem: Financial analysis is inherently complex and intimidating.

Solution:

Designed "newspaper" aesthetic for approachability Used progressive disclosure (summaries → details) Color-coded signals (green/red) for instant understanding Added onboarding steps for empty states 🎨 Design Philosophy We wanted to make professional-grade investment intelligence accessible and beautiful:

Newspaper Aesthetic: Inspired by financial newspapers (WSJ, FT)

Serif fonts for readability Border-heavy design for structure Cream/beige backgrounds for warmth Signal-First Design:

AI signals (BUY/HOLD/EXIT/WATCH) are prominent Color-coding provides instant insights Risk levels clearly displayed Story-Driven:

Every investment is linked to a story Track story evolution over time Understand the "why" behind recommendations 🔬 Technical Deep Dive AI Architecture News Articles → Entity Extraction (Gemini 3 Flash) ↓ Knowledge Graph ↓ Multi-Layer Analysis (Gemini 3 Flash) ├── Cognitive Layer (Thesis, Conviction) ├── Competitive Intelligence ├── Macro Context └── Sentiment Trends ↓ Portfolio War Room (AI Signals) Key Algorithms Story Maturity Classification

Analyzes entity count, article frequency, sentiment stability Classifies as DEVELOPING, MATURE, or ACTIONABLE Uses Gemini 3 Flash for nuanced understanding AI Signal Generation

Combines sentiment trends, risk levels, and story maturity Generates BUY/HOLD/EXIT/WATCH signals with confidence scores Provides reasoning for each signal Risk Assessment

Analyzes volatility, sector exposure, position concentration Calculates risk scores (LOW/MEDIUM/HIGH) Suggests exit strategies based on risk profile 📊 Impact & Results 28 Stories Analyzed with comprehensive AI insights 1 Active Portfolio Position with real-time AI signals 5 AI Analysis Layers providing multi-dimensional intelligence < 2 Second Response Time for complex analysis 100% Gemini 3 Flash Powered - showcasing latest AI capabilities

Built With

  • beautifulsoup4
  • fastapi
  • github
  • google-gemini-3-flash-preview
  • google-gemini-api
  • json-storage
  • knowledge-graphs
  • natural-language-processing
  • networkx
  • newsapi
  • newspaper3k
  • pydantic
  • python-3.11
  • react-19
  • react-router
  • recharts
  • render
  • scraping
  • shadcn/ui
  • tailwind-css
  • typescript
  • vercel
  • vite-7
  • web
Share this project:

Updates

posted an update —

The Financial Chronicle - Project Updates (Professional Format) Update 1: Launch Announcement Title: Introducing The Financial Chronicle - Your AI Investment War Room

Post:

Excited to announce The Financial Chronicle - an AI-powered investment intelligence platform that transforms global news into actionable insights.

What it does:

Scrapes real-time news from multiple sources Analyzes stories using Google Gemini 3 Flash Generates investment signals (BUY/HOLD/EXIT/WATCH) Manages AI-driven portfolio with risk assessment The Problem: Information overload is killing investment decisions. Thousands of articles are published daily, but which ones matter for your portfolio?

The Solution: AI that thinks like a professional analyst, connecting dots across seemingly unrelated stories to generate actionable intelligence.

Tech Stack:

Google Gemini 3 Flash Preview Python + FastAPI React 19 + TypeScript Deployed on Vercel + Render Try it live: [Your Vercel URL] GitHub: https://github.com/Harsh8818198/NewsEland

Built for the Gemini API Developer Competition.

AI #FinTech #GeminiAPI #InvestmentTech #MachineLearning

Update 2: Portfolio War Room Feature Title: New Feature: AI-Powered Portfolio War Room

Post:

Just shipped the Portfolio War Room - the flagship feature of The Financial Chronicle.

What makes it special:

AI Signal Generation Every position receives a real-time signal:

BUY - Strong bullish momentum detected HOLD - Stable position, monitor closely EXIT - Risk levels elevated, consider exit WATCH - Developing situation, stay alert Story-Driven Investing Each position is linked to the news story driving it. See exactly WHY the AI recommends each action, not just WHAT to do.

Active Alerts Real-time notifications for:

Sentiment shifts in underlying stories Risk level changes New opportunities discovered Exit triggers activated Opportunity Radar AI continuously scans 28+ stories to find emerging investment opportunities before they hit mainstream coverage.

The Technology: Powered by Gemini 3 Flash with multi-layer analysis:

Cognitive Layer (investment thesis generation) Competitive Intelligence (market positioning) Macro Context (economic trends) Sentiment Trends (momentum analysis) Risk Assessment (exit planning) Response time: Under 2 seconds for complex multi-step analysis.

Screenshot: [Add Portfolio War Room screenshot]

ProductUpdate #AIInvesting #PortfolioManagement #GeminiFlash

Update 3: Knowledge Graph Deep Dive Title: How The Financial Chronicle Connects the Dots

Post:

Behind the scenes: The Knowledge Graph Engine

Most news apps show you articles. We show you relationships.

How it works:

Step 1: Entity Extraction (Gemini 3 Flash)

Companies (Apple, Tesla, OpenAI) People (CEOs, analysts, politicians) Events (earnings, launches, regulations) Step 2: Story Evolution Tracking Stories mature over time:

DEVELOPING - Early signals, high uncertainty MATURE - Clear patterns emerging ACTIONABLE - Ready for investment decisions Step 3: Cross-Story Analysis AI finds non-obvious connections:

"OpenAI's GPT-5 launch" influences "NVIDIA earnings" "Fed rate decision" impacts "Tech stock valuations" "Regulatory changes" trigger "Sector rotation" The Result: Instead of reading 100 disconnected articles, you get ONE comprehensive intelligence report with all the connections mapped and analyzed.

Code Example:

python

Gemini 3 Flash extracts entities

entities = gemini.extract_entities(article)

Build knowledge graph

graph.add_story(story_id, entities) graph.connect_related_stories()

Track evolution

maturity = gemini.classify_maturity(story) Current Stats:

28 stories analyzed 100+ entities tracked 5 AI analysis layers Under 2 second response time

KnowledgeGraph #AI #DataScience #NetworkAnalysis

Update 4: Design Philosophy Title: Why The Financial Chronicle Looks Like a Newspaper

Post:

Design Deep Dive: The Newspaper Aesthetic

Financial data is inherently intimidating. Our goal was to make it approachable without sacrificing professionalism.

Inspiration: Classic financial newspapers (Wall Street Journal, Financial Times) - trusted, professional, highly readable.

Design Principles:

Serif Typography

Optimized for long-form content readability Professional, trustworthy appearance Newspaper-style headlines for hierarchy Color Psychology

Cream/beige backgrounds (warmth, approachability) Black borders (structure, clarity) Green = Bullish, Red = Bearish (instant understanding) Progressive Disclosure

Summaries presented first, details available on demand Scannable headlines for quick information gathering Expandable sections to prevent overwhelm Signal-First Design

AI signals are prominently displayed Risk levels clearly indicated Action items highlighted for quick decision-making Comparison: Traditional FinTech: Cold, complex, overwhelming The Financial Chronicle: Warm, clear, actionable

Technology:

Tailwind CSS for utility-first styling shadcn/ui for beautiful, accessible components Custom newspaper-style design system Screenshot: [Add UI comparison or main dashboard]

UIDesign #UXDesign #DesignSystem #FinTechDesign

Update 5: Gemini 3 Flash Integration Title: Why We Chose Gemini 3 Flash (And How We Use It)

Post:

Technical Deep Dive: Gemini 3 Flash in Production

We integrated Google's latest Gemini 3 Flash Preview across 5 distinct AI layers. Here's the technical breakdown:

Why Gemini 3 Flash?

Speed: Under 2 seconds for complex multi-step reasoning Intelligence: Understands nuanced financial contexts and relationships Cost-Effective: Production-ready pricing model Consistency: Reliable output quality with structured prompts

Our 5 AI Analysis Layers:

Layer 1: Cognitive Analysis

python

Investment thesis generation

thesis = gemini.generate_thesis(story) conviction = gemini.score_conviction(thesis) winners, losers = gemini.identify_impact(story) Layer 2: Competitive Intelligence

python

Market positioning analysis

positioning = gemini.analyze_positioning(company) advantages = gemini.find_competitive_edges(company) Layer 3: Macro Context

python

Economic trend analysis

trends = gemini.analyze_macro_trends(story) timing = gemini.assess_market_timing(story) Layer 4: Sentiment Analysis

python

Trend detection over time

sentiment = gemini.analyze_sentiment_trends(story) momentum = gemini.calculate_momentum(sentiment) Layer 5: Risk Assessment

python

Portfolio risk evaluation

risk = gemini.assess_risk(position, story) exit_strategy = gemini.plan_exit(position, risk) Key Technical Learnings:

Prompt engineering is crucial for consistent results Structured outputs (JSON) work best for integration Caching layer prevents redundant API calls Async processing dramatically improves user experience Performance Metrics:

28 stories analyzed 100% AI-powered insights Under 2 second average response time Production-ready reliability

GeminiAPI #AI #MachineLearning #ProductionAI

Update 6: Milestone Celebration Title: Hackathon Submission Complete

Post:

MILESTONE: Hackathon Submission Complete

After 9 intensive days of development, The Financial Chronicle is officially submitted to the Gemini API Developer Competition.

Project Statistics:

5,000+ lines of code written 5 AI analysis layers implemented 28 stories analyzed and tracked Under 2 second response time achieved 6 major features shipped Full-stack deployment completed Features Delivered:

Real-time news scraping from multiple sources Knowledge graph engine for entity tracking Multi-layer AI analysis system Portfolio War Room with AI signals Automated signal generation Risk assessment and exit planning Newspaper-inspired user interface Technology Stack:

Google Gemini 3 Flash Preview Python + FastAPI React 19 + TypeScript Vercel + Render deployment Key Learnings:

AI is a reasoning engine, not just a chatbot Knowledge graphs reveal hidden market patterns User experience is critical in FinTech Gemini 3 Flash is production-ready Real-time systems are challenging but rewarding Links:

Live Demo: [Your Vercel URL] GitHub: https://github.com/Harsh8818198/NewsEland Thank you to everyone who followed the development journey.

Hackathon #GeminiAPI #AI #FinTech #BuildInPublic

Update 7: Code Snippet - Entity Extraction Title: Code Example: Entity Extraction with Gemini

Post:

Technical Implementation: Entity Extraction

Here's how The Financial Chronicle uses Gemini 3 Flash to extract entities from news articles:

python import google.generativeai as genai

Initialize Gemini 3 Flash

model = genai.GenerativeModel('gemini-3-flash-preview') def extract_entities(article_text): """Extract companies, people, and events from article"""

prompt = f"""
Analyze this article and extract:
1. Companies mentioned (with tickers if available)
2. Key people (with roles)
3. Important events

Article: {article_text}

Return as JSON with structure:
{{
    "companies": [{{"name": "...", "ticker": "..."}}],
    "people": [{{"name": "...", "role": "..."}}],
    "events": ["..."]
}}
"""

response = model.generate_content(prompt)
entities = json.loads(response.text)

return entities

Example usage

article = "Apple CEO Tim Cook announced new AI features..." entities = extract_entities(article)

Result:

{

"companies": [{"name": "Apple", "ticker": "AAPL"}],

"people": [{"name": "Tim Cook", "role": "CEO"}],

"events": ["AI feature announcement"]

}

Why this approach works:

Gemini understands context (recognizes Tim Cook as Apple's CEO) Structured JSON output is easy to process programmatically Fast execution (under 1 second for most articles) Reliable entity recognition across different article formats Next step: Building the knowledge graph from these extracted entities.

CodeSnippet #Python #GeminiAPI #NLP

Update 8: Community Engagement Title: What Feature Should We Build Next?

Post:

Community Input: Future Development Roadmap

The hackathon submission is complete, but development continues. What should be the next priority?

Option A: Real-Time Market Data

Live stock price integration Interactive price charts Automated price alerts Historical performance tracking Option B: Social Sentiment Analysis

Twitter/Reddit sentiment tracking Influencer mention detection Viral trend identification Social momentum scoring Option C: Automated Trading

Execute trades based on AI signals Comprehensive backtesting engine Paper trading simulation mode Risk management automation Option D: Mobile Applications

iOS and Android native apps Push notification system Offline data access Mobile-optimized interface Share your preference in the comments.

Alternative suggestions are welcome.

Current Feature Set:

Real-time news scraping Multi-layer AI analysis Portfolio management Knowledge graph tracking Risk assessment system

ProductDevelopment #FeatureRequest #Community

  • Future Enhancements Real-Time Market Data Integration - Live stock prices and charts Social Sentiment Analysis - Twitter/Reddit sentiment tracking Automated Trading - Execute trades based on AI signals Mobile App - iOS/Android native applications Multi-Language Support - Global news in multiple languages Advanced Backtesting - Historical performance simulation

Log in or sign up for Devpost to join the conversation.