Developers

Shaffra ships as containers on Kubernetes, inside your own perimeter. Below is the sizing tool our engineers use to spec that deployment: the real component list, the real numbers, running here in your browser.

Size your infrastructure

Runs in the browser · nothing is sent anywhere

01/Infrastructure sizing

Size the deployment yourself.

This is the tool our own engineers run before a deployment, with the real component catalogue, the real per-node figures, and the same arithmetic. Choose what your deployment needs and it writes the Infrastructure Requirements Document your IT team will actually be handed.

Document header

What the IRD will be filed under. Every field is optional. Leave one blank and the document prints a neutral placeholder.

Stays in this browser

Components

Section 1Core PlatformMandatory

The Shaffra Solutions (Core) Platform is a containerized micro-services application deployed on Kubernetes. This document specifies the infrastructure requirements that need to be provided by the IT team for an on-premise deployment.

These five are the platform itself. They are always part of a deployment, so they are selected and cannot be turned off.

  • The node acts as the core of the platform, managing the Kubernetes API, scheduling workloads, and maintaining the global state of the cluster.

    CPU
    4 vCPU
    RAM
    8 GB
    Disk
    100 GB SSD (High IOPS for etcd)
    GPU
    N/A
    Qty
    × 1

    Key requirement Provision VM with Kubernetes API access & 100GB SSD for etcd.

  • These are the primary compute resources where the application microservices are deployed and executed. They handle the processing power and memory required to run the platform's logic and user-facing services.

    CPU
    4 vCPU
    RAM
    16 GB
    Disk
    100 GB SSD
    GPU
    N/A
    Qty
    × 2

    Key requirement Provision VMs for general compute; must have 10Gbps internal network.

  • Shaffra Digital Humans using Audio2Face and webgpu. Requires One GPU node.

    CPU
    16 vCPU
    RAM
    64 GB
    Disk
    100 GB SSD
    GPU
    1 x NVIDIA L4 (24 GB VRAM)
    Qty
    × 1

    Key requirement Provision VM with GPU support for real-time avatar rendering.

  • These dedicated nodes host the platform's stateful data layers (PostgreSQL), requiring prioritised disk I/O performance. Separating them from general workers prevents application traffic from interfering with critical database read/write operations.

    CPU
    4 vCPU
    RAM
    16 GB
    Disk
    200 GB SSD
    GPU
    N/A
    Qty
    × 1

    Key requirement Provision VM with NVMe/High-IOPS SSD storage (min. 5000 IOPS).

  • This is a hardened, "jump box" VM that provides the only secure entry point for administrators to access the internal cluster network. It is used strictly for SSH proxying and remote terminal access, ensuring the production nodes are never directly exposed to the wider network.

    CPU
    2 vCPU
    RAM
    4 GB
    Disk
    50 GB SSD
    GPU
    N/A
    Qty
    × 1

    Key requirement Harden OS; provide SSH access restricted to authorized Admin IPs.

Section 2Data Sovereignty

Check the self-hosted dependencies required to replace cloud APIs

  • Store User Video, Audio, Image, and Text files in self-hosted storage.

    CPU
    4 vCPU
    RAM
    16 GB
    Disk
    200 GB SSD (High IOPS)
    GPU
    N/A
    Qty
    × 1

    Key requirement Provision storage nodes for object storage with high IOPS.

    Includes Store User Video · Store User Audio · Store User Images · Store User Text

    Notes Based on the resolutions and # of users, the storage requirements will be determined for audio and video files.

  • Self-hosted Large Language Model server for inference.

    CPU
    26 vCPU (13 cores)
    RAM
    234 GB
    Disk
    200 GB SSD
    GPU
    1 x NVIDIA H100 (80 GB VRAM)
    Qty
    × 1

    Key requirement Provision VM with high RAM and GPU support for LLM inference.

    Models gpt-oss-120b

  • Self-hosted embedding model server for generating embeddings. Required when using Self-hosted LLM.

    CPU
    12 vCPU
    RAM
    85 GB
    Disk
    200 GB SSD
    GPU
    1 x NVIDIA A100 (40 GB VRAM)
    Qty
    × 1

    Key requirement Provision VM with GPU support for embedding model inference.

    Models TBD

  • Self-hosted speech-to-text service for audio processing.

    CPU
    4 vCPU
    RAM
    16 GB
    Disk
    100 GB SSD
    GPU
    1 x NVIDIA L4 (24 GB VRAM)
    Qty
    × 1

    Key requirement Provision VM with GPU support for speech recognition processing.

    Models TBD

  • Self-hosted text-to-speech service for audio generation.

    CPU
    4 vCPU
    RAM
    16 GB
    Disk
    100 GB SSD
    GPU
    1 x NVIDIA L4 (24 GB VRAM)
    Qty
    × 1

    Key requirement Provision VM with GPU support for TTS processing.

    Models TBD

Section 3Features

Additional feature components

  • Digital Humans with Livekit and BEY integration. Requires two machines, 1 for Livekit-SFU and 1 for Livekit-Agents. Requires External API Calls + Redis.

    CPU
    4 vCPU per machine (8 vCPU total)
    RAM
    16 GB per machine (32 GB total)
    Disk
    200 GB SSD
    GPU
    N/A
    Qty
    × 2

    Key requirement Provision 2 VMs, 1 for Livekit-SFU (4 vCPU, 16GB RAM) and 1 for Livekit-Agents (4 vCPU, 16GB RAM). Requires Redis and external API access.

    Depends on External API Calls · Redis

    Notes This configuration supports up to around 25 concurrent users.

Section 4Miscellaneous

Additional infrastructure options

  • Infrastructure for monitoring and logging services.

    CPU
    2 vCPU
    RAM
    4 GB
    Disk
    100 GB SSD
    GPU
    N/A
    Qty
    × 1

    Key requirement Provision monitoring stack (Prometheus, Grafana, etc.).

  • High availability configuration for critical components.

    CPU
    Additional nodes
    RAM
    Additional nodes
    Disk
    Additional nodes
    GPU
    N/A
    Qty
    Multi-node

    Key requirement Configure HA for control plane and database nodes.

    No fixed figure Sized as additional nodes rather than a fixed figure, so it adds nothing to the totals above. The count depends on which components you choose to duplicate. Confirm it with the Shaffra team.

  • This layer acts as the single entry point for all external traffic, providing a stable IP address to route user requests into the cluster. It handles SSL/TLS termination and ensures traffic is balanced across available healthy nodes. It can also be handled by ingress-nginx in Kubernetes.

    CPU
    4 vCPU
    RAM
    16 GB
    Disk
    100 GB SSD
    GPU
    N/A
    Qty
    × 1

    Key requirement Provide 1 Static IP Address and open Ports 80/443. Alternatively, use ingress-nginx in Kubernetes.

The output

Infrastructure Requirements Document

Document ID
SHF-IRD-DRAFT
Version
1.0
Date
Not set
Client
Your organisation
Prepared by
Your name
Classification
Confidential

1. Purpose

The Shaffra Solutions (Core) Platform is a containerized micro-services application deployed on Kubernetes. This document specifies the infrastructure requirements that need to be provided by the IT team for an on-premise deployment.

2. Aggregate requirements

vCPU
34
RAM
124GB
Storage
650GB
GPU
1

Totals are the sum of each selected component's first stated figure multiplied by its quantity. They describe the cluster you have selected, not a quotation.

3. Selected components

Section 1 - Core Platform

  • 1.1Control Plane Node

    4 vCPU · 8 GB · 100 GB SSD (High IOPS for etcd) · no GPU · × 1

    Provision VM with Kubernetes API access & 100GB SSD for etcd.

  • 1.2Worker Nodes

    4 vCPU · 16 GB · 100 GB SSD · no GPU · × 2

    Provision VMs for general compute; must have 10Gbps internal network.

  • 1.3Worker Nodes (GPU)

    16 vCPU · 64 GB · 100 GB SSD · 1 x NVIDIA L4 (24 GB VRAM) · × 1

    Provision VM with GPU support for real-time avatar rendering.

  • 1.4Database Nodes

    4 vCPU · 16 GB · 200 GB SSD · no GPU · × 1

    Provision VM with NVMe/High-IOPS SSD storage (min. 5000 IOPS).

  • 1.5Management & Bastion Host

    2 vCPU · 4 GB · 50 GB SSD · no GPU · × 1

    Harden OS; provide SSH access restricted to authorized Admin IPs.

4. Provisioning checklist

  • 1.1 Provision VM with Kubernetes API access & 100GB SSD for etcd.
  • 1.2 Provision VMs for general compute; must have 10Gbps internal network.
  • 1.3 Provision VM with GPU support for real-time avatar rendering.
  • 1.4 Provision VM with NVMe/High-IOPS SSD storage (min. 5000 IOPS).
  • 1.5 Harden OS; provide SSH access restricted to authorized Admin IPs.

5. Assumptions

  • All nodes run on the operating systems stated per component, on a private network inside the client's own perimeter.
  • Storage figures are the boot and working volumes of each node. Retained media, backups and log retention are sized separately against expected volume.
  • Figures follow the current Shaffra reference architecture and are indicative until confirmed against a workload profile.

Generated in your browser · shaffra.com/developers

02/The models

The models

  • shaffra-tts-2Text to speechMost of the world's languages, trained further on Gulf and Levantine Arabic, MSA and Azerbaijani. 24 dialects, ~210 ms.
  • shaffra-stt-2Speech to textStreaming transcription for Gulf dialects, MSA, English and Azerbaijani. The ears of every Shaffra employee on a phone.
  • shaffra-tajweedRecitationThe in-house Quran recitation model, 41 tajweed rules, in training now.
  • Large Action ModelsActionModels that learn a procedure by doing it rather than reading about it. Seven patents in progress.

03/Already connected

Already connected

The platform speaks to the systems your organisation already runs. An AI employee arrives inside them, not beside them.

GmailOutlookTwilioWhatsAppGoogle CalendarSAPOracleSalesforceHubSpotSlackMicrosoft TeamsZendeskJiraStripeZoomServiceNow

Bring us your numbers.

Send the IRD you just generated and we will walk through it, covering concurrency, retention, what to duplicate for high availability, and the sandbox credentials to build against.

Talk to the team