Skip to contentEmail Sergio
Back to home
Available web appboltlit.com

BoltLit: making difficult reading easier to navigate.

The problem

Understanding every sentence is not the same as following the argument.

Dense reading breaks down in a specific way. You understand every sentence separately and still lose the argument. The help you need is usually one tab away, which means it arrives too late to be useful.

Intended users. Students working through material that is above their current comfort level, and self directed readers going through papers, textbooks, and long form arguments without a class around them.

An illustration of attention drifting across a long passage. Synthetic illustration. Not recorded brain data.

Demonstration

A short passage, and an explanation on request.

Illustrative reading interaction

A continuous signal has a value at every instant, but a computer can only keep a finite list of numbers. Sampling is the decision of how often to look. Look often enough and the list can stand in for the original wave. Look too rarely and two very different waves become the same list, and nothing later in the analysis can tell them apart.

Plain language: imagine photographing a spinning wheel. Take pictures fast and the motion is obvious. Take them slowly and the wheel can appear to turn backwards, or not at all. Sampling rate is how fast you take the pictures, and once the photos are taken, no amount of clever editing recovers the motion you missed.

Written for this site. Not live AI and not the BoltLit backend.

Design

Decisions I made and why.

01

Support next to the source, not instead of it

An explanation that replaces the paragraph teaches you the explanation. The original text stays primary and everything else is secondary in weight and position.

02

Small units of help

Long generated summaries invite skimming. Shorter units, requested deliberately, keep the reader in the text.

03

Structure as a first class object

Argument maps exist because losing the shape of an argument is a different failure than not knowing a word, and it needs a different tool.

Disclosure

What I built, what I would test, what it cannot claim.

What I built+
  • Designed and built the web app end to end
  • Designed the reading surface so support sits next to the text instead of replacing it
  • Built the feature set described below and shipped it publicly at boltlit.com
  • Feature in the app: Contextual explanations placed close to the passage being read
  • Feature in the app: Notes attached to what you are reading
  • Feature in the app: Argument maps for tracking how a piece of writing is structured
  • Feature in the app: Reading progress across a document
What I would test next+
  • A simple comparison task with and without contextual explanations, scored on later recall rather than on how the reading felt.
  • Instrumenting where readers stop and reread, since that is closer to the behaviour I actually care about.
  • Asking whether argument maps change what readers remember about structure specifically.
Limitations+
  • I have not run any study on whether BoltLit improves comprehension or retention. I do not claim that it does.
  • The feature list is a description of what the app does, not evidence that it works.
  • There is no adoption, revenue, or research validation to report.
  • Support of this kind can create a feeling of understanding that outpaces actual understanding. That risk is untested here.

Research connection

Building reading tools made me curious about how attention and understanding can be measured. That is a motivation for wanting to work in a lab, not a completed neuroscience study.