Live session management
OperationalLecturers create sessions with a course code, subject, introduction and summary. Each session gets a six-character join code and a QR code, and moves through Scheduled → Live → Paused → Ended.
Our Technology
A technical overview for university IT teams, technology partners and funders. It covers what each part of the system does, how the parts connect, and which features are operational and which are still in development.
Lecturer microphone
Lecturer's voice in the lecture hall
Audio capture
Browser captures short audio chunks
Speech-to-text
AI254 speech recognition service
English transcript
Sentence-level English text
Kiswahili translation
AI254 translation pipeline
Terminology processing
Subject glossary applied and linked
Student display
Phones, tablets, laptops
Capture is on: each audio chunk moves through recognition, translation and terminology processing before it reaches students.
Captures the lecturer's voice. A clip-on or headset microphone connected to the lecturer's laptop gives the clearest input in a lecture hall.
The lecturer's browser records short audio chunks and sends them to AI254 LIVE only while capture is switched on. Switching it off stops audio leaving the device straight away.
The AI254 speech recognition service converts each chunk of audio into English text.
Text is assembled into sentence-level statements, given a sequence number within the session and stored so students who join late can catch up.
Each statement is sent to the AI254 translation pipeline together with the session's subject.
Glossary entries for the subject are passed to translation and linked to the statement, so students can open the definition of each term.
Students' devices receive new statements as they are published and show English and Kiswahili together, or Kiswahili on its own.
System components
Lecturers create sessions with a course code, subject, introduction and summary. Each session gets a six-character join code and a QR code, and moves through Scheduled → Live → Paused → Ended.
The lecturer's browser captures audio with the standard MediaRecorder API, applying echo cancellation and noise suppression where the device supports them. Chunks are forwarded server-side to the speech service; the web application does not keep audio files.
Provided by the AI254 speech recognition service through a single adapter module. The interface does not depend on any particular recognition engine, so it can be improved or replaced without rebuilding the application.
Provided by the AI254 translation pipeline. Each request includes the subject and the relevant glossary entries; the response is stored with the English statement and the processing time reported by the service.
A managed glossary of English terms, Kiswahili terms, definitions in both languages and usage notes. Each entry carries a review status (Draft, In review, Reviewed), so it is clear which terms have been approved.
Students join with their admission number, without creating an account. They can read recent statements, open terminology explanations, write notes kept on their own device and download them, ask questions by voice in English or Kiswahili, and leave or rejoin.
Sign-in, session creation for any STEM subject, start/pause/end, microphone capture on/off, permission for students to ask spoken questions, a live view of transcription and translation, connected-student count and incoming questions.
Administrators can see every session and its activity, manage lecturer accounts, maintain terminology, review demonstration requests and contact messages, and read system and error logs.
Processing time for recognition and translation is recorded for every statement and shown in the administration logs. These measurements are used to guide improvements. We do not publish latency or accuracy guarantees.
Staff passwords are stored as salted hashes, sign-in sessions use signed, HTTP-only cookies, and administration endpoints check the user's role. Students are identified only by admission number. IP addresses are stored only as salted hashes for spam and rate-limit checks. Credentials are kept in server environment variables and never sent to the browser.
“Operational” means the feature works in the current AI254 LIVE application. “Connects to AI254 services” means AI254 LIVE passes the work to AI254's speech and translation services through a defined integration point. Where those services are not connected, AI254 LIVE runs in a clearly labelled demonstration mode with sample lecture content.
Roadmap
These come directly from what the classroom pilots showed and from our technical evaluation.
Shortening the time between the lecturer finishing a statement and the Kiswahili text appearing on students' screens.
Improving how reliably speech recognition picks up specialised scientific terms.
Better segmentation when lecturers speak quickly, correct themselves or leave sentences unfinished.
Clearer layouts, text sizes and update behaviour on small phone screens in bright lecture halls.
Extending the reviewed glossary beyond the initial Biology work to other STEM subjects.
Try AI254 LIVE yourself, or ask us for a technical walkthrough for your IT team.