BandUp · AI-Powered IELTS Mock Testing Platform
Web-based IELTS mock testing platform with AI-driven evaluation, automated scoring for writing & speaking, and instant feedback for individuals and coaching centers.
Full-Stack · FYP
AI/ML
Personal / FYP
5 tech
TL;DR
“BandUp is my Final Year Project: a web-based IELTS mock testing platform with AI-driven evaluation and instant feedback. NLP and speech-processing models score the Writing and Speaking sections, while a structured test-prep flow tracks progress over time. Designed for both individual learners and coaching centers.”
Problem & Approach
The story of why this project existed and the shape of the solution.
BandUp is built around a simple bet: most candidates can practice IELTS structure, but feedback is the bottleneck. The platform lets a candidate take a full mock test (Reading, Listening, Writing, Speaking), then closes the loop with automated scoring and targeted feedback, without waiting for a human grader.
The Writing module evaluates task achievement, coherence, lexical resource, and grammatical range using a fine-tuned NLP pipeline. The Speaking module pipes audio through a speech-to-text layer and then evaluates pronunciation, fluency, and vocabulary range against IELTS band descriptors.
For coaching centers, an admin surface aggregates progress across cohorts, turning individual mocks into a teaching signal. The architecture keeps the model-serving layer isolated behind FastAPI, with a Next.js front-end that streams scored results back as soon as a section is graded.
Stack
Backend
- FastAPI
AI / ML
- LangChain
- Hugging Face
Frontend
- Next.js
- Tailwind CSS
Features
- 01Full IELTS mock test flow: Reading, Listening, Writing, Speaking, in one session.
- 02AI-driven scoring for Writing (task response, coherence, lexical, grammar) and Speaking (fluency, pronunciation, vocabulary).
- 03Instant feedback with per-criterion breakdowns instead of opaque band scores.
- 04Coaching-center mode: cohort progress, per-student histories, weak-area heatmaps.
- 05Structured test prep: practice runs that build into a calibrated band prediction.
Challenges & Refactors
Aligning automated scoring with human IELTS examiner judgments: naive language-model scoring drifted from real band descriptors.
Anchored the scoring prompts and fine-tuning data on the published IELTS band descriptors, then calibrated against a small set of human-graded sample essays.
Audio quality varied wildly across devices, hurting Speaking module accuracy.
Added pre-processing (noise reduction, silence trimming, sample-rate normalization) before transcription, and a confidence threshold that asks the candidate to retry if the transcription is uncertain.
Gallery