Assignment Assistant — the full guide

1. What Assignment Assistant is

Assignment Assistant takes an assignment question your lecturer set, and gives you back the thing you ask for: either a guided outline(structured plan + predicted mark scheme — you write the answer) or a full draft (complete prose grounded in your course materials — you edit and submit). The choice is yours per question.

Either mode rides on the same underlying engine — internally called GAIE, the Guided Assignment Intelligence Engine. It detects whether the question is theoretical or mathematical, retrieves the relevant chunks from your course library, predicts the marking scheme, and pulls lecturer pattern data. The only thing the mode changes is whether the output you see is bullets or paragraphs.

2. Guided vs Full Draft modes

On the upload page you pick a mode before submitting. The mode applies to theory questions; maths questions always come back as step-by-step working regardless of mode.

Guided outline (recommended default)

• Returns a bullet-point scaffold: intro angle, 2–4 body sections with headings + sub-points, conclusion bullets.
• Plus a predicted mark scheme — criteria, weights, and the key concepts a marker will scan for.
• Plus lecturer pattern insight when you provide a lecturer name.
• You write the actual prose. Each bullet is an incomplete phrase — a point to expand, not a finished sentence.

Full draft

• Returns a complete answer in finished academic prose — intro paragraph, 2–4 body sections with headings + full paragraphs, conclusion paragraph.
• Grounded in your uploaded course materials so it reflects what your lecturer actually teaches, not a generic web answer.
• Includes the same predicted mark scheme and lecturer pattern data alongside the draft.
• Comes with a one-click Copy button so you can paste it into your editor of choice.
• Edit it for your own voice and review before submitting — your institution's policy on AI assistance is your responsibility.

3. Plans & monthly limits

Assignment Assistant is a paid feature. Free accounts can see the upload page but need to upgrade before they can analyse anything. Limits below are monthly, counted across both file uploads and typed submissions.

There is also a per-day request limit shared with the rest of the platform — the assignmentAnalysis bucket caps you at 10 analyses per 24 hours regardless of plan, because each analysis fans out to several AI calls. If you hit the daily cap, wait — your monthly quota is untouched.

4. How to submit an assignment

Open the dashboard sidebar and choose Assignment Assistant, or land on the upload page from a course's "Upload assignment" button. You have two input modes.

Upload file mode

1. Drag a file onto the drop zone, or click to browse.
2. Accepted formats: PDF, DOC, DOCX, JPEG, PNG, WEBP. Max 10 MB.
3. Pick your institution → program → course. If you opened the page from a course, this is pre-filled and locked.
4. Optional: add the assignment title (helps you find it later) and the lecturer's name.
5. Click Upload Assignment. A progress bar shows during upload, then you are redirected to the result page.

Type manually mode

1. Click the Type Manually tab at the top of the upload page.
2. Paste or type the assignment content into the textarea. Minimum is 50 characters — short enough for a single question, long enough that one-word inputs are rejected.
3. The character counter at the bottom right turns green once you cross the minimum.
4. Pick the course and (optionally) title and lecturer, then click Submit Assignment.

5. Theory vs Mathematical routing

The first thing GAIE does is decide which path your question belongs on. It scans for equations (=, algebraic patterns, fractions, powers, integrals), calculus and trig functions (sin, cos, ∫, d/dx), matrices, and mathematical keywords like solve, calculate, derive, integrate, simplify, factorise. Numeric density also matters — three or more numbers in close succession tips the scale.

• MATHEMATICAL — runs the deterministic Math Solver Engine, then asks the AI to explain each step in plain English.
• THEORY — runs the five-layer guided academic engine described below.

If you want to check the routing without using a credit, the platform exposes a quick type-detect check used internally for UI pre-rendering — it returns the question type with no AI call.

6. Theory path — the five layers

For non-maths questions, GAIE assembles your result by running five layers. Layers 1 and 2 run in parallel; then 3, 4, and 5 run in parallel after.

Layer 1 — Question Analysis

Reads your question and works out:

• Bloom's level — remember, understand, apply, analyze, evaluate, or create. Higher levels expect longer, more argued answers.
• Structure type — definition, explanation, comparison, analysis, problem-solving, evaluation, essay, or listing.
• Action verbs the question is using (e.g. discuss, evaluate, compare).
• Core topics & sub-topics — pulled out by AI so the rest of the pipeline knows what to ground against.
• Estimated word count — 150 words for a remember question, up to 800 for create.

Layer 2 — Source Retrieval

GAIE searches your course's lecture chunks using semantic similarity (pgvector). It grabs the top matches, throws away anything below a 0.70 similarity score, and computes an internal confidencevalue — how well your uploaded course materials actually cover this question.

If internal confidence is low (no good chunks in your course library), the result page will flag it so you know the outline is leaning on general academic knowledge rather than your specific lecturer's notes.

Layer 3 — Structured Outline

This is the headline output — a bullet-point scaffold for your answer:

• Intro — 3 bullets covering definitions, angle, and roadmap.
• Body — 2–4 sections, each with a heading, 2–5 bullets, and a Bloom's hint telling you what cognitive level that section should hit.
• Conclusion — 3 bullets covering restatement, implication, and final thought.
• Word guide — suggested total length to hit.

Each bullet is an incomplete phrase — a point to expand, not a finished sentence. That is deliberate. The outline is a thinking tool, not a draft.

Layer 4 — Marking Prediction

GAIE plays examiner. It returns:

• 3–5 marking criteria, each with a weight (summing to 1.0), key terms, and the Bloom's level required to satisfy it.
• Total marks — usually 10, 15, 20, or 25, matching typical Nigerian university marking sheets.
• Key concepts must-have — 3–6 concepts a marker will scan for. If yours doesn't mention them, expect to lose marks.
• Focus areas — 2–4 constructive things to watch for while writing (not a list of mistakes — guidance).

Layer 5 — Lecturer Pattern Insight

Only runs if you provided a lecturer name. Pulls together:

• High-frequency topics for this lecturer in this course.
• Overlap between their handouts and their past questions — high overlap means "what they teach is what they test".
• Likely focus areas — the union of frequent topics and high-yield overlap.
• Plain-language summary — "This lecturer's exam questions are very closely / moderately / somewhat aligned with their handouts. Their most tested topics are X, Y, Z."

7. Mathematical path — the solver

For maths questions, GAIE routes to the Math Solver Engine. It first tries to classify the problem into one of these categories:

• Linear equations
• Quadratic equations
• Systems of equations
• Differentiation
• Integration
• Simplification
• Evaluation
• General (AI-only)

For the deterministic categories (linear, quadratic, simplification, evaluation), the solver computes the answer itself using a maths library — no hallucinations possible. For everything else it falls back to an AI-guided path. Either way, the result you see is the same shape:

• Step-by-step solution — numbered steps, each with what was done, the expression at that step, and the reasoning.
• Final answer.
• Formulas used — power rule, quadratic formula, integration by parts, etc.
• Marking insight — what a marker would scan for (e.g. "show the substitution step before the final value").
• Focus areas — sign errors, units, when to factorise vs use the formula.
• Exam tip — one practical pointer for problems of this shape.

8. Image uploads & OCR

Scanned questions or photos of handouts go through the QIP pipeline — Query Intake Processor. It runs OCR on the image, cleans up the extracted text (equation symbols, line breaks, common OCR errors), detects the domain (algebra, calculus, statistics, etc.), and then routes through the solver if it is a maths question.

• Accepted formats for image analysis: JPEG, PNG, WEBP, TIFF, BMP. Max 10 MB.
• The result includes the original OCR text (so you can spot mis-reads), the cleaned problem, and the same step-by-step solution + marking insight a typed question would get.
• Image quality matters — a well-lit, straight, in-focus photo gives clean OCR. A blurry phone shot at an angle gives noise.

9. Reading the result page

After submission you land on /assignments/result/[id]. The page polls every 3 seconds while processing, then settles into one of four states: Pending → Processing → Completed (or Failed).

Once Completed, you'll see (depending on the assignment type and what was extracted):

• Score card — if a grade was inferred, with a coloured progress bar (green ≥70%, amber ≥50%, red below).
• Exam Overlap — what percentage of the topics in your assignment have shown up in past exam papers for this course. Higher = the assignment is essentially exam revision in disguise.
• Predicted Carryover — probability that these topics appear in your next exam. Treat this as a study-priority signal.
• Topics Detected — the list of topics GAIE extracted from your submission, rendered as pills.
• View Course Insights button — jumps you into the deeper analytics for the course this assignment belongs to.

10. Lecturer pattern insight

The lecturer insight is the layer most students underuse. It only fires when you fill in the optional lecturer name on the upload form. Once it does, CampusTutor cross-references the lecturer's uploaded handouts against their past question papers and computes:

• Concentration score — 0 to 1. Above 0.75 = this lecturer reuses their handout topics heavily. Below 0.50 = they pull from outside material; you can't just memorise their slides.
• High-yield topics — the topics that appear in both their handouts and their past papers. These are statistical near-certainties for the next exam.
• Top tested topics — what this lecturer historically tests most often.

Use this to weight your revision. Two topics on the same outline are not equally important if one of them is on the lecturer's high-yield list and the other is filler.

11. How source grounding works

Theory analyses are grounded against your course's lecture chunks — the same content base your AI Tutor sees. When you upload course materials (PDFs, slides, handouts) in Tutor Mode, those documents are split into chunks, embedded, and stored in a vector index per course.

When you submit an assignment for that course, GAIE embeds your question and pulls the top semantically similar chunks. Only chunks with a similarity score of ≥ 0.70 count. The mean of the top results becomes the internal confidence score.

What happens when the question isn't in your handout

Nigerian lecturers often set assignments on topics outside their handout — extension reading, current-affairs questions, applications of theory you'll cover later. When that happens (internal confidence drops below the 0.70 bar), GAIE doesn't silently fall back to Claude's general training. Instead, it reaches out to public academic sources:

• Wikipedia — for broad topic coverage: definitions, named theories, frameworks, historical context. We pull plain-text article extracts.
• OpenAlex — for academic depth: peer-reviewed paper abstracts in your topic area. Free, open-access database with 240M+ works.

These external chunks are mixed with whatever internal chunks were found (if any), and the analysis layers (outline, draft, marking prediction) get told to hedge their language — "based on standard treatments of this topic" instead of phrasing that implies your lecturer covered it.

On the result page, you'll see a grounding bannerlisting exactly which external sources were used. That way you know whether the analysis came from your course library, the public web, or both — and you can cross-check key claims against your lecturer's emphasis before submitting.

12. Choosing a mode per question

You pick the mode per submission, so the question itself can drive the choice. A few rules of thumb that work for most Nigerian tertiary students:

Pick Guided when…

• The assignment is a topic you'll be tested on in the exam — writing it yourself is the revision.
• You have time. Guided + your own writing is slower but produces better understanding and a less detectable submission.
• You want to learn the structure of academic argument in your discipline.
• The course is one where your lecturer's personal style matters — your prose, shaped by the mark-scheme insight, will fit better than generated prose.

Pick Full Draft when…

• You're stuck and need a starting point — even if you rewrite half of it, the structure gives you traction.
• The assignment is low-stakes filler in a non-core course and you need to clear it fast.
• You've already mastered the material and just need the typing done.
• English isn't your strongest writing language and you want a baseline of academic prose to edit into your voice.

How the engine handles each mode

Both modes run the same five-layer pipeline. Layer 3 is the only thing that changes — guided returns the bullet outline, draft returns finished prose instead. Layers 1, 2, 4, and 5 (question analysis, source retrieval, marking prediction, lecturer pattern) are identical, so you get the same predicted mark scheme and lecturer pattern card in both modes.

13. How to get maximum results

Submit one question at a time

GAIE analyses one question per submission. A five-question past paper should be five separate submissions. Don't paste the whole paper in — the engine will treat it as one giant question and the outline will be vague.

Always include the lecturer name

Without it, Layer 5 is skipped entirely. With it, you get exam-pattern intelligence that turns a generic outline into a course-specific revision plan.

Build your Tutor Mode library before you need it

Internal confidence below 0.70 means your outline is leaning on general academic knowledge. To push it above 0.70, upload your lecturer's slides, past handouts, and your own notes to the course's materials. Do this at the start of the semester so every assignment you submit benefits.

Sanity-check the OCR before trusting an image result

On image submissions, the first thing to read is the cleaned problem text — confirm it matches the printed question. If a digit, sign, or variable got mis-read, the rest of the working is wrong even if it looks beautiful.

Use the marking prediction as a self-check, not a checklist

After you've written your answer, open the marking prediction again and ask: did I hit each criterion at the Bloom's level it asked for?If the criterion required analyzeand your paragraph just describes, that's a re-write before submission.

Re-run the same question after course-material updates

If you upload significant new lecture notes mid-term, re-running an earlier question will often produce a tighter outline because the source-retrieval layer now finds better chunks. (Each re-run costs one analysis from your monthly quota.)

Treat the carryover percentage as a study signal

On the result page, a high Predicted Carryovermeans these topics are statistically likely to show up in your next exam. Prioritise revision time on the high-carryover assignments — they're free practice for what's about to be tested.

14. Troubleshooting

"You are not enrolled in this course"

The course you picked isn't linked to your account yet. Open the course from the dashboard once — that creates the enrolment record — then come back.

Upload sits at 100% then fails

File made it to the server but processing failed. Check the file type and size. PDFs over 10 MB or password-protected PDFs are the usual culprits. Try switching to Type Manually mode and paste the text in.

Result page says "Processing" forever

Polling stops at 60 seconds. After that, just refresh — the analysis is still running on the server. Most theory analyses finish in 30–60 seconds; image OCR + math solver can take a little longer.

Outline looks generic

Three common causes. First, the lecturer field was empty — re-submit with it filled in. Second, your course has no uploaded materials, so the external fallback had to do all the work. Upload your lecture PDFs to Tutor Mode, then re-run. Third, the topic is genuinely off-syllabus — check the grounding banner on the result page; if it lists Wikipedia/OpenAlex sources, the analysis is running on public sources, not your lecturer's emphasis. Cross-check key claims before submitting.

I see a "Grounded from public sources" banner

That's the external-RAG fallback firing. It means your course library didn't cover this question, so GAIE pulled supplementary context from Wikipedia + OpenAlex. The analysis is still useful, but it reflects general academic conventions rather than your lecturer's specific take. The banner lists exactly which external sources were used so you can verify anything important before writing your answer.

Math solver returned "general" category

The problem didn't match a deterministic pattern (linear, quadratic, etc.), so the engine used the AI-guided path. The steps are still useful — but double-check the arithmetic, because no symbolic solver verified them.

OCR mis-read the question

Retake the photo with better lighting and the page held flat. If it's a scanned PDF, switch to Type Manually mode and paste the cleanest version of the question text.

"Monthly assignment analysis limit reached"

You've hit your plan's monthly cap. Either wait for the quota to reset at the start of the next month, or upgrade — Pro raises the cap from 30 to 150, Graduate to 300.

15. FAQ

Will the AI write my assignment for me?

Only if you ask it to. In Guided mode you get bullets and a mark scheme — you do the writing. In Full Draft mode you get finished prose grounded in your course materials, ready to edit. The math path always returns step-by-step working regardless of mode. The choice is yours per question.

Can lecturers detect that I used Assignment Assistant?

It depends entirely on what you submit. If you used Guided mode and wrote the answer in your own words, there's nothing AI-shaped for a detector to flag. If you submit a Full Draft verbatim, AI detectors and an attentive marker can often spot it — generated prose has stylistic fingerprints. The safer move with Full Draft is to use it as a starting point, restructure paragraphs, swap in your own examples and phrasing, then submit.

Does it work in any subject?

Yes — the engine doesn't hard-code any discipline. The theory path handles anything from law to literature to engineering theory. The math path covers calculus, algebra, statistics, engineering maths, and most numerical work. For very specialised symbolic work (advanced abstract algebra, formal proofs), you'll get an AI-only path with no deterministic verification.

What languages are supported?

English. The engine is calibrated for Nigerian tertiary education conventions — phrasing, marking sheet styles, common departmental expectations. It understands UK and Nigerian spellings (analyse, factorise) interchangeably with American forms.

Can I edit my submission after uploading?

Not directly — each submission is its own analysis. If you spot a problem, submit a corrected version. That counts as a separate analysis against your monthly quota.

What happens to my uploaded files?

Files are stored against your account and linked to the course. They're never used to train AI models. Other students — including those at your institution — cannot see your submissions.

Why does my analysis show different criteria than my lecturer's mark sheet?

The marking prediction is GAIE's best guess based on the question structure and (if available) historical marking patterns at Nigerian universities. It isn't connected to your lecturer's actual scheme. Use it as a strong heuristic, not as the official rubric.

Can I use it during an exam?

Don't. CampusTutor isn't an exam-room tool. Most institutions classify any AI-assisted answer during a closed exam as academic misconduct. Use it for assignments, take-home tests with explicit AI-allowed policies, and self-revision.