Use Case Intelligence RSS Search use cases Search
NVIDIA Software / SaaS Indonesia

Rafiqspace.ai

Document Processing · Security and Compliance

Rafiqspace.ai needed to deliver compliance-grade Bahasa Indonesia transcription and speaker attribution for government and regulated-enterprise meetings, but manual transcription was slow and inaccurate and commercial speech tools were not tuned for Indonesia's regulatory vocabulary or multi-speaker environments.

Read the original story Explore all use cases
1.65x real-time streaming
2.3% word error rate
569x real-time batch transcription
90% reduction in per-hour transcription cost

Reported impact

Business outcome

Rafiqspace.ai achieved 97.7 percent Bahasa Indonesia transcription accuracy, reached 569x real-time batch transcription, cut per-hour transcription cost by 90 percent, reduced manual effort by 95 percent, and enabled same-day meeting records for public-sector clients that previously waited days.

  • Compliance
  • Cost Reduction
  • Productivity Improvement

Solution

Rafiqspace.ai built a fine-tuned speech-intelligence platform on NVIDIA Nemotron models and NVIDIA accelerated computing to handle transcription, diarization, and post-processing for official meeting records. The company fine-tuned the Parakeet-CTC ASR model on a proprietary dataset covering parliamentary proceedings, internal meetings, lectures, and broadcast talk shows using NVIDIA H100 GPUs. In production, NVIDIA L40S GPUs run inference either on premises or in hybrid deployments with GCP and AWS. Speaker attribution is handled with NVIDIA NeMo ClusterDiarizer, which combines Silero VAD, TitaNet-Large embeddings, and two-stage agglomerative clustering, while bidirectional gRPC streaming delivers sub-300 millisecond interim transcription results and a punctuation-capitalization model converts raw output into readable records. The result is an end-to-end Indonesian speech pipeline built specifically for government compliance and enterprise documentation needs.

Data flow

Meeting audio from parliamentary, health, and enterprise sessions is streamed into the fine-tuned Parakeet ASR service, H100-trained models and L40S inference generate transcriptions, ClusterDiarizer assigns speakers, and a punctuation-capitalization model converts the raw output into structured meeting records delivered in real time or batch mode.

Solution architecture

5 components · 2 layers
  1. Compute
    • NVIDIA H100 GPUs Provide the training and fine-tuning compute used to adapt the speech models to Bahasa Indonesia.
    • NVIDIA L40S GPUs Power production inference for on-premises and hybrid government deployments.
    • NVIDIA NeMo ClusterDiarizer Handles speaker diarization and attribution for multi-speaker meeting records.
  2. Serving
    • NVIDIA Nemotron Provides the underlying speech model family Rafiqspace.ai uses for Bahasa Indonesia transcription.
    • Parakeet-CTC ASR Acts as the fine-tuned automatic speech-recognition model for government meeting transcription.

Architecture clues

  • Bidirectional gRPC streaming is used to keep interim transcription latency below 300 milliseconds.
  • Production inference can run on premises or in hybrid form while sensitive records remain in controlled environments.
  • The diarization stack combines Silero VAD, TitaNet-Large speaker embeddings, and two-stage agglomerative clustering.

Evidence from the source

Bidirectional gRPC streaming for sub-300 ms interim transcription results.
By fine-tuning an open NVIDIA Nemotron™ speech, Parakeet ASR model on NVIDIA H100 and L40S GPUs, Rafiqspace.ai achieved 97.7% transcription accuracy, or 2.3% word error rate (WER).
NVIDIA L40S GPUs for production inference, deployed on premise or in hybrid configurations with GCP and AWS.
NVIDIA NeMo™ ClusterDiarizer for speaker diarization, combining Silero VAD, TitaNet-Large speaker embeddings, and two-stage agglomerative clustering.
Per-hour transcription cost reduced from up to $50 (outsourced) to approximately $5—a 90% reduction.
Transparency Evidence & methodology 100/100 reference value · 89% confidence · 1 source

Review how this reference was scored and trace each normalized claim back to its documented source.

01

Scoring methodology

How scores are calculated

Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.

Customer reference value100/100
  • Problem clarity20/20
    Measures whether the customer's operational problem is specific enough to compare.
  • Implementation depth25/25
    Rewards named components, data flow, integrations, and architecture detail.
  • Outcome specificity20/20
    Measures whether the reported benefit is concrete rather than a generic claim.
  • Quantified impact25/25
    6 metrics describe measurable impact.
  • Deployment scale10/10
    Rewards evidence of a live or scaled deployment and meaningful integration scope.
Evidence confidence89%
  • Source authority20/25
    The main evidence is an official vendor-published customer story.
  • Claim traceability21/25
    10 of 12 key fields link to a named source.
  • Direct evidence20/20
    5 direct source quotes retained.
  • Date provenance10/10
    The publication date comes from the source page or publisher.
  • Source transparency8/10
    1 source reference document publisher, URL, authority, retrieval, and supported fields.
  • Implementation specificity10/10
    Rewards source-backed components, data flow, quantified impact, and a specific outcome.
02

Traceability

Sources & provenance

Field coverage shows which facts each source supports. Dates are shown only when documented by the source.

  1. NVIDIA

    Rafiqspace.ai customer story

    First-party
    https://www.nvidia.com/en-us/case-studies/transcription-accuracy-with-nemotron/
    Source type
    Vendor-published customer story
    Published
    Jun 23, 2026
    Supports
    • Customer
    • Business problem
    • Solution
    • Products
    • Solution components
    • Data flow
    • Integrations
    • technical area
    • use case category
    • Business outcome
    • outcome category
    • Metrics
    • Architecture
    • evidence quotes
    • Publication date

Date evidence

  • Published on source page Jun 23, 2026 · NVIDIA · First-party
    Open date source
Not separately sourced Industry, Region, Company size. Treat these normalized attributes as lower-confidence until a dedicated source is added.

Related

Similar use cases

See all like this
Google Cloud Archived · Sep 17, 2026 Parsewise Software / SaaS United Kingdom For risk specialists in the financial services and insurance industries, reading can feel like a full-time job. Whether assessing an insurance claim or approving a loan, these subject matter experts must search through… 29% lower error rate than best-performing LLM on grounded-reasoning benchmark Document ProcessingRisk Management Gemini · Gemini 3.0 Flash · Gemini 3.5 Flash Google Cloud Modified · Jul 13, 2026 UJJI AI Software / SaaS United Kingdom Before the strategic partnership between UJJI AI and Google Cloud, the development of a single corporate training, especially for banks and financial entities, struggled with slow and linear production cycles, requiring… , guaranteeing 99.9% availability Document ProcessingIT Operations Automation Cloud Run · Gemini · Gemini Enterprise NVIDIA Jul 7, 2026 French Ministry of Economy and Finance Public Sector France French public servants needed to answer citizen requests against more than 100 million documents per year while maintaining legal accuracy, security, confidentiality, and compliance, but manual retrieval and analysis of… 1,000 work hours saved per month for a typical 100-user deployment Document ProcessingFraud Detection DeepBrain · NVIDIA AI Enterprise · NVIDIA DGX platform