In government institutions, meetings are not just discussions.
They become records, references, decisions, and follow-up actions. Every word matters. Every speaker matters. Every delay in documentation can affect how quickly teams review, edit, and distribute official meeting results.
For the E-Notulen Expert Staff at the institution’s Secretariat, meeting documentation is closely connected to speed, accuracy, and efficiency. The team does not only need to capture what was discussed in a meeting. They also need to manage an end-to-end workflow that turns long, complex, multi-speaker conversations into clear, structured, and usable meeting summaries.
In their testimony, the team shared that the use of AI for meeting results at the institution has been highly beneficial. It helps speed up work, save time, improve efficiency, and support the editing process. They also highlighted that the wording and text accuracy produced by the system are already strong enough to help improve the quality of meeting summaries.
This is where Rafiqspace.ai comes in.
Through its Meeting and Regulatory Analysis solution, Rafiqspace.ai helps public-sector teams transform complex meeting audio into a faster, more accurate, and more structured documentation workflow.
Rafiqspace.ai is an Indonesian AI software company developing enterprise-grade and responsible Bahasa Indonesia speech intelligence for government agencies, national institutions, and regulated enterprises. Its solution is designed for high-stakes meeting environments where transcription accuracy, speaker identification, speed, and data control are critical.
The Challenge: Government Meeting Workflows Need More Than Basic Transcription
Government meetings are complex.
They often involve multiple speakers, formal language, institutional terms, regulatory vocabulary, and long discussion flows. In provincial and national legislative environments, meeting documentation requires not only accurate transcription, but also clear speaker attribution, reliable wording, and a workflow that supports review, editing, and official distribution.
Before AI-powered transcription, manual documentation could take significant time. Manual transcription for parliamentary sessions previously delivered around 70–80% accuracy and required 3–4 hours of human effort per hour of audio, making same-day distribution of official records difficult.
Commercial speech-to-text tools were also not enough. Many were not specifically fine-tuned for Bahasa Indonesia regulatory vocabulary or multi-speaker government settings. Streaming transcription performance could also fall below real-time, making live note-taking difficult in practice.
For the institution, the need was clear: a system that could strengthen the end-to-end meeting workflow, from speech capture and speaker identification to transcript review, editing, and meeting summary preparation.
The Solution: Bahasa Indonesia Speech AI Built with NVIDIA Technology
To solve this challenge, Rafiqspace.ai built a fine-tuned Bahasa Indonesia Automatic Speech Recognition and speaker diarization platform using NVIDIA technology.
The platform uses NVIDIA Nemotron speech models and Parakeet ASR, fine-tuned with Rafiqspace.ai’s proprietary Bahasa Indonesia datasets. These datasets include parliamentary proceedings, Google FLEURS benchmarks, internal business meetings, lectures, and broadcast talk shows.
From the infrastructure side, Rafiqspace.ai uses NVIDIA H100 GPUs for model training and fine-tuning, and NVIDIA L40S GPUs for production inference. The system also uses NeMo ClusterDiarizer for speaker diarization, bi-directional gRPC streaming for sub-300ms interim transcription results, and punctuation-capitalization post-processing to turn raw transcription into structured, readable meeting records.
For sensitive government deployments, Rafiqspace.ai partnered with Lintasarta, NVIDIA’s only Cloud Partner in Indonesia, to support on-premise infrastructure. This allows sensitive parliamentary and institutional records to remain in controlled environments.
In simple terms, Rafiqspace.ai did not only build a transcription tool.
It built a speech intelligence platform designed for the real conditions of Indonesian government and enterprise meeting workflows.
The Impact: Faster Workflow, Better Accuracy, and More Efficient Editing
The result is a significant improvement in how meeting documentation workflows can be managed.
Rafiqspace.ai achieved 97.7% Bahasa Indonesia transcription accuracy, equal to 2.3% word error rate. For government meeting documentation specifically, the platform reached 95.57% accuracy, compared with 91.78% for Google Gemini Pro.
The speed improvement is also substantial. The platform reached 569x real-time batch transcription on NVIDIA H100, increasing from a previous 6–10x baseline. Its streaming performance also improved to 1.65x real-time, enabling live note-taking that can better keep pace with meeting discussions.
For documentation teams, the practical impact is clear. Manual transcription effort was reduced by 95%, from 2–3 hours per audio hour to under 5 minutes with full speaker identification. Meeting minutes that previously required 2–3 days can now be delivered on the same day.
Cost efficiency also improved. Per-hour transcription cost was reduced from up to USD 50 for outsourced transcription to approximately USD 5, representing up to 90% cost reduction.
But beyond the numbers, the most important impact is felt by the users.
For the documentation team at the institution, AI helps make the editing process faster and more efficient. Instead of spending most of the time reconstructing meeting content manually, the team can focus more on reviewing, editing, and improving the quality of meeting summaries.
As the team shared, the use of AI for meeting results at the institution has significantly improved work efficiency, saved time, and supported the editing process because the text and wording are accurate and helpful for meeting summaries.
Why This Matters for Public-Sector AI in Indonesia
This success story shows that AI adoption in government is not only about using advanced technology.
It is about solving real operational problems.
For public-sector institutions, regulated enterprises, and national organizations, meeting documentation is a critical workflow. It affects accountability, compliance, institutional memory, and decision follow-up.
Rafiqspace.ai’s work with NVIDIA technology shows how Bahasa Indonesia AI can be built for local institutional needs, not only generic global use cases. The platform is designed around the language, context, vocabulary, and compliance needs of Indonesian organizations.
It also shows that AI works best when it supports people, not replaces them.
In this case, AI helps note-takers, documentation teams, and compliance officers work faster and more accurately. Human review remains important, but the repetitive and time-consuming part of transcription becomes significantly lighter.
What Comes Next
Rafiqspace.ai is now expanding beyond transcription into agentic compliance automation. The company is evaluating NVIDIA Nemotron 3 Nano for LLM-powered regulatory insight and compliance report generation, while also planning to integrate NVIDIA AI Enterprise and NVIDIA Run:ai for infrastructure optimization as deployments scale.
Its roadmap includes end-to-end compliance automation, regulatory insight generation, industry-specific models for finance, legal, and healthcare, B2C services in 2027, Indonesian regional languages, and expansion into ASEAN markets such as Malaysia, Thailand, and the Philippines.
For Indonesia, this is an important direction.
The future of AI will not only be about models that understand language. It will be about AI systems that understand local institutions, local workflows, and local responsibilities.
Through its NVIDIA-powered Bahasa Indonesia speech intelligence platform, Rafiqspace.ai is helping show what that future can look like: AI that works with people, strengthens institutional workflows, improves operational efficiency, and helps government and enterprise organizations move from manual documentation cycles to faster, more reliable meeting outcomes.