Truthline
Don't trust one AI option - trust the disagreement
Created on 5th September 2026
•
Truthline
Don't trust one AI option - trust the disagreement
The problem Truthline solves
Most AI fact-checkers rely on a single model's judgment and present it as the answer, even though any one model can be confidently wrong. Truthline solves this by cross-verifying every claim across three independent models running on GonkaRouter, and explicitly flagging when they disagree instead of quietly averaging that disagreement away. People can use it to check a claim they read online, paste a full article URL to have it automatically broken into individual factual claims and checked one by one, or verify claims in different languages — making it easier and safer to spot misinformation than trusting a single AI's opinion.
Challenges we ran into
The biggest hurdle was a chain of environment-configuration issues rather than logic bugs. Locally, Vercel's Python builder kept defaulting to an old Python 3.11 install even after we installed a newer version, which conflicted with a dependency requiring Python 3.12+ — fixed by clearing Vercel's build cache so it picked up the correct interpreter. We also hit a subtle bug where an environment variable name and its actual secret value got swapped in the code, causing a KeyError that looked like a missing-key problem but was actually a one-line logic mistake. On deployment, a folder named with capital letters (API instead of api) caused Vercel to silently 404 on our backend route, since routing requires an exact lowercase match. Each issue took careful log-reading (build logs vs. runtime logs vs. local terminal output) to isolate, since the same symptom (500/404 errors) had several unrelated root causes.
Tracks Applied (1)
AI For Society
Gonka Router
Technologies used