Tuturaksara
- Year
- April 20, 2026
- Role
- AI Engineer
- Client
- PT. Mitra Pinasthika Mulia
- Access
- Internal only
- Speech-to-text
- LangChain
- vLLM
- Node.js
Internal system — no public link or screenshots.
What follows describes the problem, the approach and the engineering decisions at a level that discloses no proprietary detail.
Tuturaksara converts recorded audio into text, and then into a finished document, most often the minutes of a meeting.
What it does
- Transcription: audio in, timestamped text out.
- Templated output: the transcript is rendered into a document structure such as minutes of meeting, rather than handed over as a raw wall of text.
- Feeds downstream tools: its transcripts are one of the inputs to Rewangit, which covers the wider document-generation case.
Approach
Transcription and document assembly are separate stages. Keeping them apart means the transcript remains a verifiable record of what was actually said, while the generated minutes are a derived artefact that can be regenerated or reformatted without re-running the audio.