What We Do
Speech, language and classroom work, developed together
AI254’s work covers the full path of a lecture: capturing what the lecturer says, translating it with the correct scientific meaning, delivering it to students during the lecture, and checking how well it works.
Real-time lecture transcription
During a lecture, the lecturer's speech is captured from a microphone and converted into English text in short sections, so students see each statement within moments of it being spoken. The transcription keeps sentence boundaries, so each statement can be translated as a complete thought rather than as loose fragments.
A lecture hall is a hard place for speech recognition. The main challenges we work on are:
Classroom noise
Movement, conversation, fans and the room’s echo all reach the microphone along with the lecturer’s voice.
Rapid speech
Lecturers speed up, repeat themselves, correct themselves and leave sentences unfinished.
Specialised vocabulary
Words such as ‘endoplasmic reticulum’ are rare in everyday speech, so general models often get them wrong.
English-to-Kiswahili lecture translation
Each completed English statement goes through the translation pipeline, which produces Kiswahili text for students. Translation uses the subject being taught: when a session is set to Biology, the Biology terminology resources guide how technical terms are rendered.
Preserving scientific meaning matters more than producing a literal sentence. A translation can be fluent and still be wrong if it changes what a process does or confuses two related concepts. Keeping terminology consistent matters as well: if ‘osmosis’ is rendered one way in the first lecture and another way in the fifth, students lose the thread that connects the course.
The English statement always stays on screen next to the Kiswahili, so students can check one against the other and nothing the lecturer said is hidden.
Scientific terminology development
We work with Kiswahili language specialists and subject experts on university-level Biology vocabulary. Each glossary entry includes the English term, the agreed Kiswahili term, definitions in both languages and notes on usage. Entries go through review before they are treated as approved.
Scientific terminology needs interpretation in context. Swapping one word for another is not enough. Some examples:
English (in a Biology lecture)
Direct substitution
Contextual rendering
cell
chumba · simu
seli
General tools may choose ‘room’ or ‘mobile phone’. In Biology the word always refers to the basic unit of life.
cellular respiration
kupumua kwa seli (breathing)
upumuaji wa seli
Respiration in Biology is a chemical process that releases energy. It is not breathing, and the translation has to keep that difference.
the powerhouse of the cell
nyumba ya nguvu ya seli
kituo cha nishati cha seli
A teaching metaphor. Translated literally it means nothing, so it is rendered by meaning: an ‘energy station’.
selectively permeable
inayopenyeka kwa kuchagua
yenye upenyezaji teule
Describes a property of the membrane. It needs one consistent term, used the same way in every lecture.
Examples are for illustration. Approved terms come from the reviewed AI254 Biology glossary.
Classroom technology
AI254 LIVE provides three interfaces that share one live session:
Student interface
Students join with their admission number and a session code or QR code. They read the English transcription and Kiswahili translation, open terminology explanations, keep notes and ask questions by voice or text.
Lecturer interface
Lecturers sign in, create a session for their subject, start and stop the live session, switch microphone capture on or off, check the output and see how many students are connected.
Administrator interface
Authorised staff manage lecturer accounts, monitor sessions, review system logs and errors, maintain terminology resources and handle demonstration requests.
Open AI254 LIVE to try the student and lecturer experience in demonstration mode.
Testing and evaluation
We test AI254 in real lectures and record what we find. The two classroom pilots at Maasai Mara University used several sources of evidence:
Student questionnaires
Students’ experience of following the lecture, their comprehension, participation and confidence.
Lecturer interviews
How the system fits into normal teaching, and what lecturers noticed in the output.
Classroom observations
How students used their devices during the lecture and where problems came up.
System logs
Processing times, recognition and translation errors, and connection behaviour during sessions.
Findings go straight back into development, especially work on technical vocabulary, translation delay and readability on mobile.
AI254 is an evolving technology
AI254 LIVE is still under active development and has been tested in early-stage pilots only. Wider deployment across courses or institutions will need further testing, evaluation in other teaching environments, and expanded terminology resources.
See the work in practice
Request a demonstration to see AI254 LIVE in a lecture setting and talk through how it could be evaluated in your department.