Guide

GemmaLens turns academic papers, technical documents, and subtitles into source-grounded language lessons. The goal is not to replace reading with translation; it is to help you read the next source with less support.

Paper reading

Start with the original source. GemmaLens opens the first prepared lesson quickly, then continues preparing the rest while you read.

Step 1

Upload or paste

PDF, DOCX, markdown, text, or transcript becomes a source document. PDFs keep the original file when available.

Step 2

Read beside the source

For PDFs, the original page stays visible while the learning panel explains the current section.

Step 3

Read while it prepares

The first page appears immediately. By default, GemmaLens prepares every page section in the background to reduce waiting.

Step 4

Build the paper map

As sections finish, GemmaLens builds an outline of argument flow, concepts, terms, and expressions.

Step 5

Save what matters

Save concepts, terms, expressions, and hard sentence patterns separately so review can stay targeted.

Local-first preparation

GemmaLens does not wait for a whole-paper pass before becoming useful. It breaks long sources into small jobs, which lets local Gemma models prepare useful lessons without sending your paper to a remote reading service.

ThinkPad / small local model

Runs the same section pipeline with lower memory pressure; background preparation hides per-section delay.

Mac M1 Max

Fast local inference can prepare many sections while the learner reads the visible page.

Mobile / edge demo

The product story still works on constrained devices because work is split into small page-section jobs.

Learning objects

GemmaLens separates different kinds of help so the lesson does not become a wall of generated text.

Key ideas

Ideas needed to follow the paper's argument, such as the method, objective, ablation, or benchmark setup.

Terms

Source-grounded vocabulary and domain terms worth saving only if they are unfamiliar or repeated.

Reusable expressions

Academic moves such as contrast, limitation, method setup, result claims, and conclusion language.

Sentence patterns

Dense structures are simplified and explained so the learner can read similar sentences later.

Translation

Use Translate for short passages. Use Documents or Video for real study, because those workspaces preserve source context, repeated terms, and review history.

Video study

Video mode treats subtitles as timestamped reading material. It is strongest when you have a reliable English subtitle file or stable public transcript.

Local subtitle study

Open a local video and attach English subtitles. The timeline follows playback and supports line-level study.

Live cues

While watching, GemmaLens surfaces lightweight vocabulary and phrases from the current subtitle window.

Scene lessons

Analyze a short scene for concepts, spoken expressions, and source-grounded review items.

Recap watched part

Use deeper recap after watching a larger segment, then save useful terms and expressions for review.

Reading levels

B1

Needs help with main ideas, academic phrases, and dense grammar.

B2

Can read general academic text with support for domain words and long sentences.

C1

Can read research writing but benefits from structure, nuance, and reusable expression notes.

C2

Focuses on precision, rhetoric, field-specific phrasing, and paper-level argument flow.

Domain-heavy

Difficulty comes mainly from specialist concepts rather than grammar.

Notes

PDF pages remain the visual source. Extracted text powers the lesson, so complex equations or unusual columns may need source-side checking.
The paper map is built from section lessons instead of one large summary, which keeps local inference responsive.
Translation is for quick sentence help. Documents and Video are the main learning workspaces.
Video learning works best with verified subtitles, either local SRT/VTT files or a stable public transcript.