Meetpal Bahasa Indonesia Speech and Meeting Intelligence

Rafiqspace.ai (Rafiq Space Intelligence) is an Indonesian AI software company building enterprise-grade speech intelligence for Bahasa Indonesia. It serves government agencies, national health institutions and regulated enterprises, where meetings, parliamentary sessions and compliance reviews must be documented accurately. The company also runs Threatpal for security analytics, Edupal for language learning and Taskpal for task management on a shared agent runtime.

Meetpal transcribes live or uploaded meetings and produces speaker-identified notes with audit trails, automated summaries cross-referenced against Indonesian regulation and an agent that teams can query. The speech engine fine-tunes NVIDIA's open Parakeet-CTC model, adds NeMo-based speaker diarization, streaming transcription with interim results under 300 ms and a punctuation and capitalisation model. It runs on Lintasarta, NVIDIA's cloud partner in Indonesia, and capabilities are also exposed as agent tools through an MCP catalog.

• Bahasa Indonesia ASR: fine-tuned Parakeet-CTC model for local speech and terminology.
• Speaker diarization: identifies each speaker in meetings and hearings.
• Streaming transcription: interim results in under 300 ms for live sessions.
• Regulation-aware summaries: automated minutes cross-referenced against Indonesian regulation.
• Audit trails and team agent: traceable notes that teams can query in natural language.
• MCP catalog: exposes speech and summary capabilities as standard agent tools.

Vietnamese companies with Indonesian subsidiaries or partners can transcribe and summarise Bahasa Indonesia meetings and compliance reviews. Regional banks and insurers can document Indonesian board and audit meetings with speaker attribution. The approach also offers Vietnamese public agencies a reference model for local-language meeting minutes.

The NVIDIA case study reports 97.7% transcription accuracy on the company's test set and 95.57% on government meeting documentation versus 91.78% for Google Gemini Pro. Manual effort falls from 2 to 3 hours per audio hour to under 5 minutes, and cost per audio hour drops by up to 90%, from up to 50 USD to about 5 USD. Batch transcription runs at 569 times real time on H100 GPUs.

Governments and regulated industries in Southeast Asia want sovereign speech AI tuned to local languages and hosted in-country, because global models are less accurate on local speech and data residency rules are tightening. Fine-tuning open speech models on local GPU clouds is becoming a common route for regional startups.

Rafiqspace.ai is an Indonesian company operating as Rafiq Space Intelligence. It lists Google Cloud, NVIDIA, Alibaba and NextUnicorn as partners and runs on Lintasarta infrastructure. Its roadmap includes industry models for finance, legal and healthcare and expansion to Malaysia, Thailand and the Philippines. Founding year and address are not stated on the website.