Shipping Earnings Analysis Tool

The Problem

Every quarter, shipping companies publish earnings transcripts. These documents contain valuable signals about fleet utilization, rate expectations, and strategic direction—but extracting insights is tedious work. Analysts spend hours reading through transcripts, trying to piece together what management actually said versus what they're implying.

I wanted a tool that could answer questions like:

• "What did Euronav say about VLCC rates in Q3 2024?"
• "How does Star Bulk's outlook compare to Golden Ocean's?"
• "Is this CEO dodging the question about charter coverage?"

The Approach

This tool uses Retrieval-Augmented Generation (RAG) to make earnings transcripts searchable and queryable. Instead of keyword matching, it understands context and can synthesize answers from multiple sources.

Current Architecture

1. Transcript Processing: Earnings call PDFs are parsed, chunked, and embedded using OpenAI's embedding model
2. Vector Storage: Chunks are stored in a vector database for semantic search
3. Query Engine: Questions are matched to relevant chunks, then an LLM synthesizes the answer with citations

Planned Features

Dodge Score: A metric for measuring management transparency. When an analyst asks about charter rates and the CEO pivots to talking about ESG initiatives, that's a dodge. The score will quantify this across companies.

Cross-Company Comparison: Surface contradictory statements automatically. If one tanker CEO says rates are strengthening while another says they're softening, that's worth knowing.

Management Scorecard: Track predictions against outcomes. Did the CFO's rate guidance prove accurate? Build a credibility index over time.

Technical Details

• Stack: Python, LangChain, Pinecone, OpenAI
• Data: Public earnings transcripts from SEC filings and company websites
• Processing: Batch ingestion with incremental updates as new transcripts release

Status

Currently in active development. The RAG Q&A system works locally with a handful of transcripts. Next steps:

• Expand transcript coverage to all major public shipping companies
• Build web interface for queries
• Implement dodge score algorithm

This is the project I'm most focused on right now. Updates will be posted in the Writing section as I make progress.