Inspiration
Important decisions are often made from scattered emails, notes, reports, and call records. Most summarizers flatten everything into one confident narrative. That hides a crucial difference: a documented event is not the same as someone’s claim, and neither is the same as an inference.
SourceLine was built to keep those distinctions visible.
What it does
SourceLine turns unstructured records into a chronological evidence map. It:
- extracts naturally written dates and sorts events into a timeline
- classifies each record as a fact, attributed claim, inference, or unknown
- preserves a line-level source locator for every event
- flags missing dates, assumed years, and unsupported claims for human review
- generates a copyable plain-text report
- processes everything locally in the browser
How we built it
SourceLine is a lightweight client-side web application built with semantic HTML, responsive CSS, and vanilla JavaScript. Its transparent classification engine uses explicit linguistic signals rather than a hidden black box. Date parsing, chronological sorting, evidence labels, review flags, and report generation all run on the user’s device. No account, API key, backend, or data upload is required.
Challenges
The core challenge was avoiding false certainty. A sentence can contain a date and still describe only what someone claimed. SourceLine therefore treats attribution, uncertainty language, and inference language as separate signals. It also exposes the original line reference so users can verify every result instead of trusting an unexplained summary.
What we learned
Trustworthy analysis is not just about producing an answer. It is about showing the basis for that answer. A smaller, explainable system can be more useful than a larger opaque one when people need to review evidence carefully.
What's next
Future versions could add local file import, configurable classification policies, collaborative review, export formats, and an optional on-device language model while preserving source traceability and privacy.
Log in or sign up for Devpost to join the conversation.