What is Gadfly?
Gadfly is an open source civic transparency tool built to hold the United States' Congress accountable. Meaning, it exists to clearly show Americans, irrespective of party affiliation, socioeconomic status, or class if their representatives actually represent them.How it works
Enter your zip code and Gadfly finds your House representative and two senators. For each representative, Gadfly analyzes their voting records across 11 policy areas, each split into a directional binary (e.g. Category: Economy & Cost of Living, Directional Binary: Expanding Spending / Stimulus vs Cutting Spending / Austerity) and converts it into one easily digestible voter profile. Policy areas like healthcare, housing, individual rights, and government accountability were selected for their impact on the daily lives of the majority of Americans. A sample of legislation voted on in the 119th Congress was analyzed using Claude's Sonnet 4.6 to determine which categories it touches and which directions it moves. The votes of each member of the chamber that voted on the legislation had the relevant categories and directions applied to their voter profile.Where Gadfly Is & Where Gadfly Is Going
Gadfly is currently in the proof of concept phase. The concept that it proves: congressional voting data can be compiled and organized in a way that makes it easy for everyday citizens to determine for themselves if their representatives are voting in their interest. That being said, the categorization carried out by Claude has not been verified for every house vote, and some categorizations that have been reviewed were found to be dubious. Voter profiles in their current form are to show the overall concept. Current categorization of voter records should be viewed with healthy skepticism.The data pipeline for senate voting data is currently being built. Once that is built and a sample of senate votes have been categorized the next step will be to improve the categorization process. I believe that the growing concern over large language models, and more specifically the companies building them, is not unfounded. Because of this, it is my intention to build Gadfly's own neural network that will initially categorize legislation and be followed by a human run verification process.
Open Source
Gadfly is an open source project. Contributions, data sharing, and feature suggestions are welcome.Method Behind the Madness
- House Voting Data Obtained from Congress Api - Congress API
- Member & Vote Data - All congressional member information and voting records are pulled directly from the official Congress.gov API and stored locally so the app can quickly look up your representatives and their votes.
- AI Analysis - The full text of each bill is sent to Claude, Anthropic's AI, along with a detailed analysis prompt. Claude reads the bill and returns a structured breakdown of which categories it touches, which direction it moves, and whether any red flags are present.
- Temperature set to 0 - Claude gives the most consistent, deterministic answer possible. No creativity, no randomness.
- Independent analysis - Each bill is analyzed against 11 categories and 9 flags independently.
- Large bill handling - Bills too large to process in one call are split into sections, analyzed separately, and synthesized into a single result.
- Long term goal - Future Feature: A user-friendly database giving citizens direct access to every vote, bill text, and flag, fully searchable.
- Human verification - Future Feature: A rigorous review process where verified human checks confirm AI categorizations, with clear indication to human verified votes.
Take a Look at the Prompts
See the prompts that power Gadfly