Collect and label multimodal data
Gather meeting recordings, transcripts, slide images and optional screen recordings, then label a validation set with time-stamped action items and responsible speakers.
01 / THE IDEA
Meeting action-item extraction in practice.
Action-item extraction from meeting video is a multi-modal task combining audio (ASR), video frames, slides and transcripts to produce time-stamped, attributable tasks. Key challenges include speaker diarisation, imperfect ASR, implicit or multi-part action items, and privacy/consent for recorded meetings.
Real-world use case · Proposed extension
A proposed application for product and engineering teams: highlight candidate action items after a meeting so a human can confirm assignments and deadlines. The starter code does not implement this extension.
02 / THE BUILD
How to build a meeting video action-item extractor project
step by step.
Create preprocessing pipelines
Run ASR, speaker diarisation and OCR on frames; align modalities by timestamp and extract candidate text spans for labelling.
Build a baseline extractor
Train a simple classifier or sequence labeller on transcript segments (with timestamp and speaker features) and fuse visual cues where available.
Evaluate and create review UI
Measure precision/recall for action-item extraction, inspect missed items and false positives, and present candidates for human verification.
03 / SOURCE SNAPSHOT
What the meeting video action-item extractor starter actually does.
The supplied starter loads the scikit-learn digits dataset, splits it, trains a RandomForestClassifier on digit feature vectors, prints a classification report, and shows a confusion matrix. It does not process video, audio, transcripts or extract action items.
SOURCE EXCERPTPython
model = RandomForestClassifier(n_estimators=300, random_state=42)
model.fit(X_train, y_train)
predictions = model.predict(X_test)
print(PROJECT_NAME)
print(classification_report(y_test, predictions))
ConfusionMatrixDisplay.from_predictions(y_test, predictions, cmap="Blues")Important limitation
The starter uses the sklearn digits image dataset and a RandomForest classifier for a digit-recognition baseline; there is no multimodal input (audio, transcript, slide images), no timestamp alignment, and no action-item labelling. This code cannot validate meeting action-item extraction without substantial replacement of data and model components.
04 / BEYOND THE CLASSROOM
Where a meeting video action-item extractor project could go next.
PROPOSED BUSINESS CONCEPT
Action-item extractor: a pilot concept
A proposed extension: build a pipeline that runs ASR and speaker diarisation, extracts candidate sentences with timestamps, uses a lightweight classifier or sequence labeller to tag action items, and links each item to a speaker and meeting time for human review.
WHO IT COULD SERVEEnterprise teams
FIRST MVPProduce a time-stamped list of candidate action items from meeting recordings using ASR plus a simple classifier, with a UI for a human to accept, edit or reject each item.
KEY RISKRecording meetings raises privacy and consent concerns; transcription errors and diarisation failures cause missed or misattributed action items, so a human reviewer should validate assignments before they are acted on.
This is a potential application pathway, not a delivered product. The supplied starter does not implement this service or validate it with real users.
05 / WHAT YOU TAKE AWAY
What a meeting video action-item extractor project puts on your resume.
- Design a multimodal data collection plan combining ASR, speaker diarisation and frame OCR for meetings.
- Implement preprocessing that aligns transcript segments with video timestamps and speaker identity.
- Train and evaluate a baseline action-item classifier using precision and recall at the segment level.
- Explain why the supplied starter using the digits dataset does not implement meeting action-item extraction.
Want to build this with mentors and a lab?
AI Engineer Course Networkers Home runs this topic inside its AI Engineer Course, with placement support and hands-on labs.
06 / QUESTIONS STUDENTS ASK
Meeting Video Action-Item Extractor project FAQs.
- Is this suitable as a final year project for CSE students?
- Yes — it is an advanced, multidisciplinary final year project idea involving audio, video and NLP components; expect work on modelling, data alignment and evaluation.
- What skills and tools should I learn before starting?
- Gain practical experience in Python, ASR tools, speaker diarisation, transcript processing, multimodal feature fusion and basic ML/NLP libraries like scikit-learn and PyTorch.
- How should I evaluate progress and how long will it take?
- Evaluate via metrics for extraction accuracy, timestamp alignment and speaker attribution on annotated meeting samples; a solid prototype typically takes 3–6 months depending on scope.