Engineering lectures run at slide speed. A derivation that took your teacher twenty minutes at the board becomes four lines on slide 23 — and the step between line two and line three, the one where everything happened, is not on the slide at all. Whether you are at BUET, KUET, RUET, CUET or a private university, the class test is next week and the slides are what you have.
The best use of AI for engineering students is not getting it to solve the problem set. It is getting those missing twenty minutes back. Here is the method, step by step, and the places it stops working.
Why is engineering so hard to self-study from slides?
Slides are compressed. They record what the lecturer said, not why it follows, and they assume you were paying attention at the exact moment a symbol was defined. The textbook fills the gap in theory — but Hayt, Çengel or Hibbeler run to hundreds of pages of dense English, and their notation rarely matches your lecturer's.
So the night before a term final, most students do one of two things: memorise the slides, or find a senior's solved problems and memorise those. Both work right up until a question changes one condition.
Step 1: upload the slides you were actually given
Porphy works on your own material, not a generic syllabus. Upload the lecture PDF and every study tool treats it the way it treats a course it researched itself: you can read it, ask questions about it in chat, generate practice from it, and build a concept map of it.
One limit, and it matters: the PDF must contain real, selectable text. Slides exported from PowerPoint or LaTeX usually do. A scanned handout or a phone photo of the whiteboard is an image, and there is no OCR step — so the app tells you it has nothing to read rather than guessing.
Step 2: highlight the step you can't follow
Find the line where the derivation lost you and highlight it. The live voice tutor opens and explains that exact line out loud, drawing on a shared whiteboard as it goes — the part of a lecture a PDF cannot hold. A few lines worth highlighting in a typical first- or second-year course:
- Circuits: why a capacitor behaves as an open circuit once a DC circuit reaches steady state — and an inductor as a short.
- Mechanics of materials: why the slope of the bending-moment diagram equals the shear force at that point.
- Dynamics: where the Md² in the parallel-axis theorem comes from, not just when to add it.
- Thermodynamics: why entropy is defined with dQ/T along a reversible path, and what 'reversible' is protecting you from.
- Signals and systems: what the Laplace transform actually buys you when you use it to solve a circuit.
You can interrupt it, ask it to slow down, or ask for the same idea from a different angle. It can also hand you a flashcard deck or a quick MCQ set on what you just struggled with, in the middle of the call.
Step 3: fill the prerequisite the slides assumed
Half of being stuck in engineering is being stuck one course back. Transfer functions assume the Laplace transform. Beam deflection assumes you can integrate twice and apply boundary conditions. Power systems assume phasors. This course's slides will not re-teach last semester's.
For that, give Porphy the topic instead of a PDF. It researches it on the live web and writes a course: a planner splits the topic, 5 researchers search in parallel, a critic sends thin sections back, and every part arrives with the sources it was written from. Choose one researched answer, a 3–4 part overview, a 5–7 part standard course or an 8–12 part deep dive — and tell the intake what you already know, so it starts at your gap rather than at chapter one.
Step 4: test yourself before the class test
Reading a derivation until it feels familiar is the most common way to walk into an engineering exam unprepared, because familiar is not the same as able. So test the four things separately.
Every concept in a course or an uploaded PDF is scored on memory, comprehension, structure and application, and an evaluator agent updates those scores after each practice session. 'I remember the parallel-axis formula but can't set up the problem' shows up as high memory and low application — which tells you to stop rereading and start working problems. The explain-it-back prompt is the harshest test: if you cannot say in your own words why the capacitor goes open-circuit, the tutor will tell you which part you skipped.
Where AI for engineering students stops helping
- It does not replace problem sets. Engineering is a skill of doing, and the only way to get fast at Thevenin equivalents is to find a lot of them by hand.
- It is not MATLAB, SPICE or a lab bench. It will explain what a simulation shows; it will not run one for you.
- It can be wrong, especially in arithmetic-heavy steps. Check numbers against your textbook, and claims against the sources each part lists.
- It does not know your lecturer's marking. Your department's own past papers still say more about the exam than any AI can — upload them if they are typed, and practise on them.
Thinking in Bangla, studying in English
Most engineering courses in Bangladesh are taught from English books, while plenty of students still think a hard idea through in Bangla. Porphy explains in either. In Bangla it keeps the technical terms — impedance, eigenvalue, shear force, entropy — in English, so the explanation is in the language you think in and the vocabulary still matches the slides and the exam script. How that term lock works is a post of its own.
And if you are in CSE, the same method carries over to algorithms — here is how to learn DSA with AI without letting it write your code.
Every new account starts with 2,000 credits free — no card, no subscription, and they never expire. A full researched course runs about 600–1,200 credits, an MCQ set or flashcard deck 10–30, and the voice tutor 20–60 a minute. Free accounts run a lighter pipeline so the credits go further: at most 3 parts per course, with text and tables in place of generated visualizations.