The Researcher

Home Page
How the daily news analysis looks
How the analysis is being done
How the archive show the analysis result vs actual market result

💡 Inspiration The spark for this project came from studying BlackRock's Aladdin—the legendary investment management technology that oversees trillions in assets. I was fascinated by its unique "working logic" and ability to see risks others missed.

Combining this from my own 5 years of experience in the stock market, I realized that retail traders lack this kind of "institutional brain." Most importantly, retail traders cannot process the thousands of news headlines that move markets every day. I didn't just want a tool that predicts prices based on charts; I wanted to build a News-First AI Engine that reads, understands, and reacts to information just like a human trader—but at infinite scale. When I discovered the advanced reasoning capabilities of the latest Gemini 3 Pro model, I knew I finally had the engine to build this "hardcore brain."

🧠 What it does (and our USP) The Researcher is an autonomous stock market analyst whose Unique Selling Point (USP) is its ability to synthesize Real-Time News with technical data.

While other apps show you a chart, The Researcher reads the news for you. It tracks 10 major Indian stocks and market indices (NIFTY 50), using a 6-Pillar Framework that prioritizes information flow:

News & Sentiment (Primary): Instantly analyzing breaking news, earnings reports, and global events. Social Momentum: Extracting retail buzz from social streams. Technical Analysis: Validating news impact with price action (RSI, MACD). Fundamental Anchoring: P/E ratios, ROCE, and quarterly growth. Market Regime: Adjusting strategies based on volatility. Historical Pattern Matching: Comparing current news cycles to past events. It outputs a Daily, Weekly, and Monthly prediction, each with a calibrated Confidence Score (1-10) and a detailed textual Rationale explaining why it made that call.

⚙️ How we built it The core of the system is the Gemini 3 Pro model, chosen for its superior reasoning capabilities and long context window.

Data Pipeline: We built a Python pipeline using yfinance for price data and custom scrapers for news. Context Construction: We feed Gemini a massive context window containing 200 days of price action, dozens of recent news headlines, and calculated technical indicators. Reasoning Engine ( prediction_agent.py ): This is where the magic happens. We use a structured "Chain of Thought" prompt where Gemini is forced to: Step 1: Rate the technical setup (0-10). Step 2: Rate the news sentiment (0-10). Step 3: Identify conflicts (e.g., Tech says UP, News says DOWN). Step 4: Resolve conflicts using financial logic (e.g., "Breaking news overrides weak technical divergence"). Self-Correction Loop: We built a feedback system where the AI tracks its own accuracy. If it predicts "UP" and the stock goes "DOWN", the system records this error and lowers its confidence for similar future setups. Frontend: A clean, dark-mode dashboard built with FastAPI and Jinja2 to visualize the data. 🚧 Challenges we ran into Hallucination Control: Early versions would invent news or price targets. We solved this by strictly grounding the prompt in the provided context data and implementing a "Verification Layer" that clamps predictions to realistic daily ranges (max 2-3% moves for large caps). Conflicting Signals: Teaching the AI how to weigh signals was tough. Is a Golden Cross more important than bad earnings? We used Gemini 3's advanced reasoning to create a hierarchy of rules (e.g., Governance issues > Technical Breakouts). Latency: processing extensive history took time. Gemini 3's speed improvements were crucial in getting analysis time down to seconds per stock. 🏅 Accomplishments that we're proud of The "Why" Engine: We are most proud of the Rationale generation. Seeing the AI explain, "I am predicting NEUTRAL because while RSI is bullish, the sector sentiment is negative due to recent regulatory changes," feels like talking to a real analyst. Self-Calibrating Confidence: Building the system that strictly penalizes the AI's confidence score based on historical misses. It makes the "High Confidence" badge actually mean something. 📚 What we learned We learned that Context is King. The difference between Gemini 1.5 and Gemini 3 in handling dense numerical data and subtle sentiment nuances was night and day. We also learned that for financial AI, explainability is more valuable than raw accuracy—users trust the tool 10x more when they can see the logic.

🚀 What's next for The Researcher Whole Market Analysis: Scaling our engine from 10 stocks to analyze the entire Indian Stock Market, finding opportunities in mid-cap and small-cap stocks that humans miss. Portfolio Management: Allowing users to upload their portfolios for personalized "Keep/Sell/Add" recommendations based on news risk. Building Unshakeable Trust: We aim to be the "Trust Layer" of fintech. By keeping our accuracy archives public and verifiable, we want to bridge the trust gap between retail investors and algorithmic advice.