Overview

Given a dataset of companies (companies.jsonl), the system classifies each company against three intent queries:

Query	Description
ev_battery_supply_chain	Companies in the EV battery manufacturing/supply chain
fintech_growth_proxy	High-growth fintech companies with recent funding signals
supplier_role_detection	Tier-1/Tier-2 suppliers in automotive, aerospace, or industrial sectors

Architecture

• QueryRule abstraction: Each query is defined as a QueryRule with positive evidence groups (AND across groups, OR within group), negative signals, and confidence boosters.
• Negative filtering: Explicit negative signals disqualify false positives early.
• Confidence scoring: Base confidence + booster bonuses, capped at 1.0.
• Reranker: A lightweight reranking layer resolves ambiguous multi-query matches.

Files

solution.py        # Main classifier
queries.txt        # The three intent queries
final_results.csv  # Output: company -> query matches with confidence
WRITEUP.md         # Architecture, tradeoffs, error analysis, scaling
companies.jsonl    # Input dataset (Veridion-provided)

Usage

python solution.py

Outputs final_results.csv with columns: company_name, query, confidence, matched.

Requirements

Python 3.8+, no external dependencies.

Results Summary

The classifier processes all companies in companies.jsonl and outputs per-company match results with confidence scores. Companies that match no query are recorded with matched=False.