Knowledge gap finder
"Knowledge gap finder" is a text tool that helps the teacher move from checking answers to diagnosing misunderstandings. Instead of marking "correct/incorrect" you get a breakdown of possible misunderstandings and suggestions for targeted practice. The tool works with two fields: source material (student answers, solutions, reasoning) and goal or desired result. On output - one text result that can be copied or downloaded as TXT. This is not a grading system and not a replacement for pedagogical judgment: the model may err, and service availability depends on the queue. Below is a workflow for everyday tasks with clear steps and limitations, built on a fictional example.
Open tool: Knowledge gap finder ↗What exactly the tool does
"Knowledge gap finder" accepts student text answers and matches them against the stated goal. The task is not to assign a grade, but to point out possible misunderstandings: what rules, connections or steps the student interprets differently than required. This is a diagnostic draft that saves time on initial analysis.
The interface is deliberately simple: a web form with text fields, no sliders, style presets and no settings panels. You choose the output language, press the Generate button and get one result in a separate panel. It can be copied or downloaded as plain TXT. There are no integrations, auto-save, batch processing or direct publishing. The volume limits are set in advance: the first field - up to 8000 characters, the details field - up to 1000 characters. This means that for a long test you have to select the most representative fragments, rather than inserting everything at once. Such selection is useful in itself: it forces you to formulate exactly what you want to check.
Example: Source material: "I think percentages add directly: a 20% discount, then another 10% - that's 30%." Goal: "Understand how sequential discounts are applied to the price."
Source material: "I think percentages add directly: discount 20%, then another 10% - that's 30%." Goal: "Understand how sequential discounts are applied to the price."
Preparing source material
Collect student answers in the form they were given, including intermediate steps and formulations. It is precisely the formulations that show the thought process: the phrase "I just added" says more about the model than the final number. If there are many answers, select 3-5 most characteristic ones, covering different types of errors.
Remove personal data from the text if not needed for diagnosis. The tool is not designed to store sensitive information, and the field is length-limited. Better replace names with "Student A" and keep only the educational content.
Check that the source material reads as a coherent text, not as a set of fragments. If the answer was oral and you recorded it, mark this so the model does not guess about written neatness. The context of recording helps interpret the result.
Collect 3-4 student answers on one topic, preserving intermediate reasoning. Remove names, keep only the learning content.
Formulating the goal
The goal is not "solve the problem", but a specific understanding that you are checking. Formulate it as an observable result: "Student explains why sequential discounts are multiplied, not added." The more specific the goal, the more useful the suggested gaps will be.
If there are multiple goals, split the work into several runs. One run - one goal. This keeps the result observable, and practice suggestions do not merge into a general list. Use the details field (up to 1000 characters) for clarifications: student level, covered topics, practice format.
Do not turn the goal into a list of all program requirements. The tool finds the gap between the current answer and the stated understanding; an overloaded goal blurs this gap and reduces diagnostic accuracy.
Instead of "Learn fractions", formulate "Student can explain why 1/2 + 1/3 is not 2/3 using a visual model."
Reading the result
The result usually contains a list of assumed misunderstandings and related topics for practice. Read it as hypotheses, not as a verdict. Each statement is worth checking on other student works or a short conversation: "Explain why you added percentages?"
Separate diagnosis from plan. First make sure the described misunderstanding is actually present, and only then choose exercises. If the tool indicated a gap that you do not observe, this is also a useful signal: perhaps the source material was incomplete.
Pay attention to formulations that repeat. If the same topic pops up in different runs, it is a stable candidate for targeted practice. A single mention may be a coincidence.
Possible result fragment: "Student applies additive model to sequential changes. Practice: compare 20%+10% and 0.8×0.9 on the same price, then explain the difference in words."
Organizing targeted practice
From the suggestions select two or three exercises that can be done in a short approach. Targeted practice works when it is narrow and repeatable, not when it covers the whole topic. Start with an assignment that directly pits the erroneous and correct models against each other.
Plan the understanding check in advance: ask the student to explain the solution in words, provide a counterexample, or change the condition. This will show whether the model changed, not just the answer. Without such a step it is easy to get a memorized algorithm without understanding.
Keep short notes: date, goal, identified gap, executed exercise, observation. Through several weeks these records will show dynamics better than a separate generation result. The tool gives a starting point, not a ready-made learning trajectory.
Choose two exercises: one where the student must identify the error in a peer's work, and another where they must apply the correct rule to a new price.
Limitations and typical errors
The model may err: attributing a thought to the student that was not there, or missing an non-obvious gap. The result is not a grade and does not replace a conversation with the student. Do not publish it as an official conclusion and do not make personnel or certification decisions based on it.
The tool does not support document upload, OCR, batch processing, auto-save, and export to DOCX or PDF. Only text result with copying and TXT download is available. Do not plan your workflow around features that do not exist.
Do not insert unnecessary personal data into the fields and do not rely on confidentiality beyond what is necessary. Length limits are also a reminder: diagnosis must be selective and meaningful.
If the model suggests a gap you did not see, check whether the student's original answer was fully included in the 8000-character limit.
Example workflow
Step 1. Collect 3-4 student answers on one topic, preserving intermediate reasoning. Step 2. Formulate one goal as an observed understanding. Step 3. Insert material into the first field, goal and context - into the second, select output language and run generation.
Step 4. Read the result and mark hypotheses that agree with your observations. Step 5. Check one or two hypotheses with a short conversation or additional task. Step 6. Choose two exercises and assign a check date.
Step 7. After a week repeat the run with new answers on the same goal and compare the lists of gaps. Such a cycle takes little time, but turns a one-time check into a manageable process. The scenario is fictional and given as an example of work organization.
A teacher collects answers on percentages, runs the tool, checks the suggested gap about sequential discounts, and assigns two practice exercises for the next class.
"Knowledge gap finder" is useful where you need to quickly move from checking answers to diagnosing misunderstandings. It does not grade, does not store work and does not build a learning trajectory for you. Its strong point - a structured draft of hypotheses and practice directions, which you verify with pedagogical judgment. Work with one goal per run, select representative fragments, check hypotheses with conversation and keep short notes. Remember the limitations: possible model errors, field limits, absence of export to office formats and availability dependence on the queue. With such an approach the tool saves time on routine analysis, leaving the substantive work to the teacher.
Open tool: Knowledge gap finder ↗Interface translations are machine-assisted and may need refinement.