AI Research Lab · Founded in GCC, built around the world

The AI operating layer for enterprise and government.

One platform to deploy and govern an autonomous AI workforce across your organisation. Already live across the region.

The story · 01/06

TRUSTED BY LEADING ENTERPRISES

stc
Omantel
Ooredoo
Qiwa
Ministry of Finance
MOTCIT
Tawuniya
Health Holding Company
Riyadh Airports
Sohar Port
Bahrain Chamber
Almoayyed
eXtra
solutions by stc
BTF
MJIA
MOTT
Musand

01/The platform

One platform. Four connected layers.

Layer 01

Shaffra Work

Shaffra Work · live product

The front door to the platform. A unified workspace where your teams connect with organisational knowledge, systems and AI employees, bringing every conversation and escalation into one place, with quality insights generated from every interaction.

  • One workspace for people and AI employees
  • QA on every conversation
  • Native Arabic and English

Layer 02

Knowledge Hub

Knowledge Hub · live product

The intelligence layer beneath the platform. Enterprise search, advanced RAG and knowledge graphs organise institutional knowledge, with granular access controls governing visibility and organisational memory keeping information current.

  • Advanced RAG + knowledge graphs
  • Granular access control
  • Organisational memory

Layer 03

Shaffra Build

Shaffra Build · live product

The operational engine. Connect your knowledge base, policies and systems, and configure an AI employee around them, defining personality, voice, permissions and response behaviour. Deploy in minutes with enterprise-ready controls built in.

  • Role-based knowledge access
  • Voice, dialect and avatar
  • Deploy in minutes

Layer 04 · the trust layer

Governance & Deployment

Everything above operates within it. Define roles, access levels, policies and audit controls, then decide how it scales and where it runs.

Governance PanelDeployment & Scaling
The Shaffra governance panel, showing an access-control table giving each role its access level and its scope (global admin, sales lead, ops manager, data engineer), with the last change against each, and an audit and versioning block underneath holding the full audit log and a rollback to the previous version
The Shaffra deployment panel, showing web and CRM live in production with IVR, WhatsApp and avatar still staged, a compute profile per deployment, auto-scaling rules with a session ceiling and an escalation buffer, and template replication into another department or another client
  • Role-based access and scoped permissions
  • Policies, audit logs and versioned rollback
  • Sovereign, on-premises or cloud

02/Organisational memory

Your company, connected in one graph.

  • 0

    Entities

  • 0

    Relationships

  • 0

    Sources

01 · One fact

It starts with a single known fact.

A claim number. A customer. An order. On its own, it is simply a record, the kind already duplicated across systems throughout your organisation.

02 · And the line between

The line is where the context lives.

Shaffra preserves the relationship, not just the record. This claim links to its policy, coverage, region and every interaction connected to it.

03 · The whole estate

Then every other fact your organisation holds.

Policies, tickets, contracts, call transcripts, invoices, spreadsheets. Resolved into entities, connected and kept current. Structured and unstructured, in one graph.

04 · Memory

Shaffra does not lose its memory.

Knowledge persists across sessions, workflows and AI employees. What is captured once remains governed, current and available across the workforce through the same organisational graph.

05 · Retrieval

Why was claim 88213 rejected?

Rejected under Exclusion 7B of policy P-114, for flood damage outside the endorsed period.

  • [1] Policy P-114 · clause 7B
  • [2] Adjuster note · 03.11
  • [3] Claim 88213 · loss date

03/Quick use cases

Meet your autonomous teams.

Already trained, already working, live in a minute.

Team 01 · on shift now

Project Management team

Shaffra OS · PMO · AI employee · Yara
Yara, Shaffra AI employee

PMO · AI employee

Yara

Works across the department, keeping plans, reporting and priorities aligned, with every action visible and accounted for.

OversightReports24/7

Every project and every deadline, watched

A live programme view of tasks, analysis and the decisions behind them.
A live programme view of tasks, analysis and the decisions behind them.
Shaffra OS · PM · AI employee · Ahmad
Ahmad, Shaffra AI employee

PM · AI employee

Ahmad

Runs the plan day to day: assigns work, follows progress, clears blockers and keeps every task moving to completion.

PlanningFollow-up24/7

From standups to timelines, handled

His portfolio at a glance, showing what is active, what is at risk and where action is needed.
His portfolio at a glance, showing what is active, what is at risk and where action is needed.

YARA AND AHMAD, TWO OF 8+ ON THIS TEAM.

Team 02 · MANAGING YOUR TEAM END TO END

Human Resources Team

Shaffra OS · Recruiter · AI employee · Lina
Lina, Shaffra AI employee

Recruiter · AI employee

Lina

Publishes openings, screens applicants, compares candidates against evidence and keeps the hiring process moving.

Job adsSourcingScreening

The whole top of the funnel, one employee

Her requisition rules and hiring criteria, configured once and applied consistently to every opening.
Her requisition rules and hiring criteria, configured once and applied consistently to every opening.
Shaffra OS · Interviewer · AI employee · Faris
Faris, Shaffra AI employee

Interviewer · AI employee

Faris

Reviews every CV, runs structured interviews and scores candidates against defined criteria, giving hiring teams evidence they can act on.

CV analysisLive interviewsScoring

Live: sit the interview yourself

Enter your email. Faris sends an invite and runs a structured interview, scored against defined criteria.

Free demo interview · one per email · your email is used only to send this invite.

Shaffra OS · Functional Evaluator · AI employee · Mohammed
Mohammed, Functional Evaluator · AI employee

Functional Evaluator · AI employee

Mohammed

Evaluates job readiness against role-specific competencies, tasks and real-world requirements.

HiringInternal mobilityDevelopment
Defines
The functional competencies of every job
Evaluates
People against role-specific requirements
Simulates
Role-specific tasks and scenarios
Reports
Evidence-based assessments: strengths, gaps, readiness

Lina, Faris and Mohammed, three of the 10+ on this floor.

Team 03 · answer every call

Call Center team

Shaffra OS · live line · Omar, Call Center
Omar, Shaffra's AI call-center employee

Call center · AI employee

Omar

Available now. Uses the right customer context and completes actions during and after each call, including follow-ups, tickets and service tasks.

العربيةEnglish24/7

LIVE · TALK TO HIM NOW

Enter your number. Omar calls you back within seconds, in Arabic or English.

Free demo call · one per number · your number is used for this call and nothing else.

HR team · 10+ AI AGENTSCall center · 5+ AI AGENTSPM team · 8+ AI AGENTS

One workforce. One operating layer.

Three teams shown here, part of a workforce built to scale.

Sales AgentInvestor RelationsInbound SalesVP SalesAirport AssistantPersonal EnablerCompetency EvaluatorPerformance Evaluator+ 21 more
35+
AI agents, live today
99+
on the roster by year end
5
Minutes to Deploy one into your Ecosystem

04/Shaffra Talk

Shaffra Talk · live demo
Lip sync
Locked to the phonemes it is speaking. Arabic included, not approximated.
Languages
Any language, and the accent that belongs to it. Khaleeji, Najdi, Hijazi, Egyptian, MSA and English.
Concurrency
Every conversation runs in parallel. Nobody is queued, nothing is busy, no shift ends.
Systems
Integrated with the systems you already run, including CRM, ERP, ticketing, telephony and calendars.
Surfaces
Reception desk, kiosk, web, video call, phone, and holographic.
Presence
A photoreal human avatar or rendered character, powered by Shaffra STT and TTS.
SEE SHAFFRA TALK IN ACTION BOOK A DEMO →

Talk to yourAI team, live.Face to face.

Open Shaffra Talk and speak directly with your AI team. They respond live, on screen, in your language.

What happens

No forms, queues or tickets. Say what you need and your AI employee acts while you speak, then escalates to a person when human support is needed.

BASED ON YOUR OWN PEOPLE

And the face can be one of your employees.

With permission, the avatar can reflect an employee's appearance, expressions and voice.

  1. 01Photographs

    A small set of employee images captured from multiple angles.

  2. 02Likeness

    The model learns their visual likeness and renders it consistently.

  3. 03Conversation

    The avatar responds live across supported surfaces in the caller's preferred dialect.

05/Our technology

Built in-house. Trained for the world.

01 · TTS

A voice for every language you serve.

Our TTS supports a wide range of languages, with deeper training across Gulf Arabic and MSA. Type your text, choose a voice and press Speak — the audio streams back as it is generated.

shaffra-tts-2

Streaming · 8 left

50/200

Voice

+50

Ready · Gulf Arabic

Model
shaffra-tts-2
Mode
streaming
Voices
50+

02 · STT

Speech recognition in your own dialect.

Press record and speak. Shaffra STT transcribes English, Arabic, Hindi, Mandarin and Urdu in real time, streaming each word as it is heard.

shaffra-stt-2

Streaming · live

Language

+20

Press and speak

English · one minute

0:00

Say something in English…
Live · English
Model
shaffra-stt-2
Mode
streaming
Languages
60+

03 · Avatars

A face, rendered in your browser.

Choose from three avatar approaches for a Shaffra employee. A photoreal human avatar, an in house 3D character, or a near real avatar rendered live on the visitor's GPU. Select the option that best fits the experience and deployment environment.

Shaffra photoreal human avatar speaking ArabicShaffra in-house 3D-rendered avatarShaffra real-time WebGPU avatar rendered in the browserRendering live

A real person, faithfully represented.

A machine learning avatar trained from real human footage. Speech and facial motion are synchronised for natural, frame accurate delivery.

Lip syncframe accurate
Fidelity
photoreal
Engine
generative ML model
Best for
video calls · brand

STT → reasoning → TTS → the face, rendered live in the browser

Faces
3
Render
browser
Surfaces
call · web · kiosk

04 · Subconscious AI· Flagship

Every deployment leaves the next agent better at the job.

The cognitive layer beneath every Shaffra agent. A role that has already run in production arrives understanding the work instead of learning it. The practice carries forward; your data never leaves your perimeter.

Compounding

The role arrives experienced.Every time.

Practice, carried forward
Every deployment sharpens the role.
Memory you own
Semantic and episodic, in one graph.
Your data never travels
Only the practice carries. Your records stay put.
The architecture is built on this paperAriGraph: Learning Knowledge Graph World Models with Episodic Memory for LLM AgentsAnokhin et al. · IJCAI-25Read on arXiv

Observe

01 / 04

An agent works. What it saw lands as one observation.

Facts held
41,900
Hours worked
12,480
Onboarding
inherited
Architecture
AriGraph
Memory
semantic + episodic
Onboarding
inherited

07/The start

Your next team is minutes away.

Tell us what the work is. We'll introduce you to the AI employees who already do it.

Shaping the future of work. The future starts here