Earnings-calls-search-engine

An interactive Python Dash web application that enables market research analysts to extract contextual insights from earnings call transcripts with ease and precision. The system supports keyword-based, context-rich search over parsed transcripts, powered by a custom MongoDB backend and FMP APIs for financial data integration. It is tailored for competitive intelligence and investor relations workflows.

 

🔍 Key Features

• 🔍 Automatic Synonym Expansion – Each keyword is matched against a curated synonym bank, enabling more intelligent search coverage.
• 🔍 Text Cleaning & Segmentation – Transcripts are preprocessed to distinguish between Prepared Remarks and Q&A sections, ensuring high relevance in search results.
• 🔍 Speaker Attribution – Each sentence is structured and tagged with the correct speaker name, enabling role-specific insights (e.g., CEO vs Analyst responses).
• 🔍 Contextual Search Results – Outputs are not just matches, but full-context segments that help users understand the meaning behind the mention.
• 🔍 Backend Storage – All transcripts and metadata are stored and indexed within MongoDB for fast, flexible querying.

 

🛠 Built With

• 🛠 Python Dash – for frontend interactivity
• 🛠 MongoDB – for scalable, indexed storage and advanced querying
• 🛠 FMP APIs – for integrating financial metadata
• 🛠 NLP tools – for text parsing, segmentation, and speaker tagging

 

📽️ Demo

Check out the live walkthrough of the application in action:

Earnings-search-app.mp4

 

🚀 Setup Instructions

1. Clone the Repository

git clone https://github.com/Chan-dre-yi/Earnings-calls-search-engine
cd Earnings-calls-search-engine

2. Create a Virtual Environment

python -m venv venv

Activate it (Linux/Mac):

source venv/bin/activate

Activate it (Windows):

venv\Scripts\activate

3. Install Dependencies

pip install -r requirements.txt

4. Configure the Database URI

• Host a MongoDB instance (locally or via MongoDB Atlas).
• Open db_connection.py and add your actual connection string.

MONGODB_URI = "your_mongodb_connection_string"

5. Get Your API Key

• Visit Financial Modeling Prep and obtain an API key.
• Add your key to the relevant part of the test_db_script.py script.

api_key = '' #your api key

6. Add Proxy Variables (if required)
If you're behind a firewall or using restricted internet, set your proxy variables in the test_db_script.py script like so:

os.environ["HTTP_PROXY"] = "http://your_proxy:port"
os.environ["HTTPS_PROXY"] = "http://your_proxy:port"

7. Prepare the Database
Run the 4 scripts inside the DatabaseScripts/ folder (modify the collection names if/ as required)

cd DatabaseScripts
python test_db_script.py
python doc_text_segmentation_script.py
python test_to_prod_script.py
python synonyms_generation.py
cd ..

8. Run the Project

python earnings_search_dash_app.py

 

🔌 Data Source

This project uses the Financial Modeling Prep API for accessing real-time and historical earnings call data:
Financial Modeling Prep

 

🖥️ Console View

Here’s a glimpse of the backend console showing real-time parsing and search execution:

 

🗃️ MongoDB Backend

The system relies heavily on MongoDB for:

• 🗃️ Storing cleaned and structured transcripts
• 🗃️ Segmenting text into Prepared Remarks and Q&A
• 🗃️ Expanding search queries using a synonym bank
• 🗃️ Tagging sentences with correct speaker roles

 

📌 Example Snapshots

 

❓ FAQ

❓ What problem does this solve?

• "Market research analysts often rely on secondary reports (e.g., Statista, consulting PDFs, investor decks). These rarely provide direct quotes or speaker context from earnings calls. Analysts waste hours cross-checking, verifying, and building PPTs.
  This tool makes it un-imaginably easier by combining trusted data sources with direct transcripts/quotes, letting analysts get ground truth in seconds, not days."

---

❓ Why not just use SQL instead of MongoDB?

• "Earnings call transcripts are long, unstructured text — with speakers, timestamps, sentiment, and topic tags. Modeling this in SQL would mean complex joins and performance bottlenecks.
  MongoDB allows document-based storage, making it natural to index and query by speaker, keyword, or section, with faster full-text search. It’s a perfect fit for unstructured but query-heavy data."

---

❓ How is this different from existing sources like SeekingAlpha or Yahoo Finance?

• "Platforms like SeekingAlpha, Motley Fool, Yahoo Finance provide access to earnings calls — but mostly in article format or delayed transcripts, often behind paywalls.
  Our tool provides direct, fast, and structured access: you can search who said what, where, and in which quarter instantly. It’s designed for analysts’ workflows, not general investors."

---

❓ What impact does it bring for analysts?

• "Analysts building PPTs for decision-makers usually spend days compiling trusted figures and quotable insights. This tool enables: a. Instant quote retrieval from executives’ words, b. Cross-quarter comparisons without manual digging, c. Faster, more reliable reports with traceable sources. Net impact: higher trust, less grunt work, faster insights."

---

❓ How was this project validated?

• "The stack and performance choices were constantly reviewed by ultra-senior mentors (25–30 yrs experience). The approach — MongoDB + NLP pipeline — was confirmed as the right direction for scale and usability.
  The only reason it wasn’t pushed to production: team re-org + deprioritization due to layoffs, not technical infeasibility."