Investor Sentiment Analysis Pipeline

Multi-Service Azure Pipeline for Earnings Call Analysis

An automated pipeline that analyzes earnings call transcripts and extracts investment sentiment. It uses Azure Language Service for NLP and Azure AI Foundry for synthesis, producing executive summaries of investment risks and sentiment trends over time.

Tech Stack

  • Python
  • Azure Language Service
  • Azure AI Foundry (Kimi-K2 Thinking)
  • Streamlit

How It Works

The pipeline runs in five stages:

  • Input Processing: Accepts transcripts in .txt, .pdf, or .docx format
  • Language Analysis: Azure Language Service performs sentiment analysis, key phrase extraction, and summarization
  • Data Structuring: Results saved as JSON
  • AI Synthesis: Kimi-K2 Thinking applies deeper analysis with a detailed system prompt
  • Report Generation: Markdown output with sentiment analysis and investment insights
JSON output from Azure Language Service

Demos

Terminal demo:

Streamlit interface for uploading transcripts and viewing formatted reports:

What I Learned

This was my first time chaining multiple Azure services into one workflow. Azure Language Service handles the structured NLP tasks, Azure AI Foundry adds the analytical layer. Breaking complex analysis into discrete stages (text extraction, sentiment, key phrases, synthesis) made the pipeline easier to debug and maintain.

First time deploying to Streamlit too. It made it easy to turn a command-line tool into something non-technical users could actually use.