◆ The Black Box for AI4CX & CCAI ◆
SABAI.
Multimodal · Multilingual · Multi-tenant · Agentic.
Drop into any industry. Rewire CX.
01 · The Black Box
One box.
Every channel in. Every action out.
SABAI is the Black Box for AI4CX & CCAI — drop it between your customers and your business systems, and every conversation becomes a resolved outcome.
Inputs · any channel
The
SABAI
Black Box
Multimodal · Multilingual
Multi-tenant · Agentic
Outputs · business actions
02 · The Four Pillars
Built on four non-negotiables.
Every other AI platform picks one or two. SABAI ships all four — because real CX in our markets demands it.
Multimodal
Voice, text, documents, images and APIs — one unified intelligence surface.
Single pipeline · ASR + NLU + OCR + TTS
Multilingual
Regional-first across South Asia, ASEAN and MENA — dialects, code-mixing, accents.
20+ languages live · 60+ on roadmap
Multi-tenant
Isolated workspaces, data, and models per tenant — built for BPOs, enterprises and governments.
Per-tenant KB · RBAC · audit logs
Agentic
Doesn't just answer — plans, calls tools, updates CRMs, and resolves tickets end-to-end.
Tool-use · workflows · guardrails
03 · Why Now
Four forces.
One window.
The macro stack is converging — and the next 24 months decide who owns the regional CX category.
Force 01
Foundation models commoditized
Base LLM cost ↓ 100× in 24 months
The bottom of the stack is a race to zero. Defensibility moves up — to proprietary data, regional models, and workflows.
Force 02
Voice-AI tipping point
Outperforming generic ASR on regional languages
Open ASR/TTS now beats closed English-first models on Nepali, Bhojpuri, Maithili and Sinhala when fine-tuned. The window for a voice-first CX engine built on this edge is open right now.
Force 03
Sovereign-AI mandates
DPDP · PDPL · PDP · localization laws
India, UAE, Indonesia, Nepal — all forcing in-region inference and data residency. Cloud-only US incumbents cannot comply.
Force 04
BPO economics breaking
Agent wages +8–12% YoY
Across South Asia and MENA, seat-based contact centers are losing margin while CX expectations rise. The replacement is agentic.
04 · The Shift
The next four billion
won't speak English.
From Kathmandu to Kuala Lumpur, Colombo to Cairo, Jakarta to Jeddah — they speak in Nepali, Hindi, Tamil, Sinhala, Bahasa, Javanese, Arabic, Urdu, Bengali and the dialects in between. SABAI is the intelligence layer built for that world.
4.2B
Voice-first users across South Asia, ASEAN & MENA
Source: GSMA Mobile Economy 2025 · Internet Society
60+
Official languages, thousands of dialects
Source: Ethnologue 2024
<5%
Of enterprise AI is multilingual-native
Internal analysis · top-50 enterprise AI vendors
05 · The Gap
The Intelligence Gap.
01
Customers speak in local languages. Businesses operate in English systems.
02
Regional data is unstructured — locked inside calls, PDFs, and WhatsApp voice notes.
03
Existing AI tools collapse on dialects, code-mixing, and accented speech.
04
Enterprises need privacy, sovereignty, and on-prem control — not just an API key.
06 · Our Edge
We own the global language edge.
Built from Kathmandu — grounded in the languages, dialects, and deployment realities of South Asia, ASEAN, and the Middle East. The edge global players cannot replicate from San Francisco.
Languages live in v1
South Asia
Nepal · India · Sri Lanka · Bangladesh · Pakistan
ASEAN
Indonesia · Malaysia · Singapore · Philippines · Thailand
MENA
UAE · Saudi Arabia · Egypt · Qatar · Oman
07 · The Product
Introducing
SABAI Black Box.
A multilingual, multimodal, agentic intelligence system that converts voice, text, documents, and APIs into business actions.
01
Listen
02
Understand
03
Reason
04
Act
05
Respond
08 · Inside the Box
Six layers.
One sovereign engine.
Layer 1
Input
Layer 2
Language Intelligence
Layer 3
Purpose Engine
Layer 4
Knowledge
Layer 5
Agentic Action
Layer 6
Output
09 · Trust & Sovereignty
Built for the regulated world.
Banks, telcos, hospitals and governments don't buy AI on demos — they buy on trust, residency, and control. SABAI is engineered for that buyer.
SOC 2 path
Type I → Type II in roadmap
ISO 27001
Information security management
DPDP · PDPA · GDPR
Region-aware data handling
PII auto-redaction
On voice, text & documents
Per-tenant keys
BYOK · isolated encryption
On-prem & air-gapped
No data leaves the perimeter
10 · The Revolution
Traditional CX, rewritten.
SABAI CX, redefined.
For decades, contact centers scaled by adding seats. SABAI scales intelligence — and flips the cost, quality, and language curves at the same time.
11 · Drop Into Any Vertical
One core.
Every industry.
The same Black Box powers a bank's collections agent, a telco's tier-1 support, a hospital's intake desk, and a government's citizen helpline — with multi-tenant isolation between every deployment.
SABAI
Core Engine
BFSI
Voice KYC, collections, advisory
Telco / ISP
Tier-1 support automation
Government
Citizen services in dialect
Healthcare
Patient intake & triage
Retail / D2C
Voice commerce + CX
Travel & Hospitality
Multilingual concierge
EdTech
Regional language tutors
Media
Multilingual content engine
12 · The Flywheel
The moat is the loop.
Models are temporary. Distribution is fragile. The only durable advantage in regional AI is a self-reinforcing data loop — and SABAI's compounds with every conversation.
01
More deployments
Banks, telcos, BPOs, governments
02
Proprietary corpus
Multilingual, dialectal, real-world
03
Fine-tuned regional models
Better than any global generic LLM
04
Higher accuracy · lower latency
Per language, per domain
05
Better CX outcomes
Resolution, CSAT, AHT
Every conversation makes SABAI harder to displace.
13 · The Landscape
Where everyone else isn't.
Map the market on the only two axes that matter for regional CX — and the upper-right is empty.
← Global / English-first Regional / Multilingual-native →
← Generic AI CX-native →
Regional · Generic
Regional · CX-native
Global · Generic
Global · CX-native
OpenAI
Google CCAI
AWS Connect
Cresta
Observe.AI
Talkdesk
Yellow.ai
Haptik
Gnani
SABAI
The unoccupied quadrant
Nobody else is regional-native and CX-native at the same time.
- Global generalists (OpenAI, Google, AWS) — powerful, but English-first and not built for CX workflows.
- Global CX specialists (Cresta, Observe, Talkdesk) — strong in US/EU, weak in our languages.
- Regional chatbots (Yellow, Haptik, Gnani) — multilingual, but still chat-bot era, not agentic.
- SABAI — sovereign, multilingual-native, multimodal, agentic. The only player in the upper-right.
Positioning based on public product capabilities and language coverage as of 2026.
14 · Live Demo
One engine.
Three worlds.
Voice → Intelligence
North-star SABAI pipeline
Input
Bhojpuri-Hindi mixed · ISP customer call
Black Box Output
Structured · machine-actionable
{
"language": "Bhojpuri-Hindi mixed",
"intent": "internet_not_working",
"emotion": "frustrated",
"domain": "ISP Support",
"action": "create_ticket",
"response_language": "Bhojpuri",
"sla": "P1 — 2hr"
}15 · Deployment
Deploy anywhere.
Cloud API
Private Cloud
On-Prem Box
Edge Device
Air-Gapped
16 · Business Model
One core.
Many revenue lines.
Platform
Monthly subscription
Core SABAI engine + dashboards
Usage
Per-minute voice
ASR + TTS + intent
Usage
Per-document
OCR + extraction + summary
CCAI
Per-agent / month
Live call assist + analytics
Enterprise
Annual license
Private deployment + SLA
Sovereign
On-prem license
Government / air-gapped
Setup
Implementation fee
Adapter customization
Adapters
Custom domain
Vertical-specific intelligence
Data
Model improvement pkg
Continuous fine-tuning
17 · Scope & Deliverables
What we will actually ship
in the next 6 months.
The vision is the 17 sections above. The proposal is below — concrete, scoped, and acceptance-tested. Anything not listed here is explicitly out of scope for v1.
Phase 1 · M1–M3
Foundation
SABAI Core API
ASR + NLU + intent + TTS — REST + streaming
5 languages live
Nepali, Hindi, English, Maithili, Bhojpuri
Single-tenant workspace
Auth, KB upload, conversation logs
1 vertical adapter
ISP customer support — tier-1 automation + escalation
Cloud deployment
Single-region, managed infra, 99.5% SLO
Internal admin console
Tenant config, prompts, guardrails
Acceptance
Demo with 1 internal use case end-to-end · 100 test conversations passing
Phase 2 · M4–M6
Production v1
Multi-tenant isolation
Per-tenant data, keys, RBAC, audit logs
Agentic workflows
Tool-use, CRM write-back, ticket resolution
Live agent assist (CCAI)
Real-time suggestions, call summaries, QA scoring
+5 languages
Bengali, Tamil, Sinhala, Bahasa, Urdu (10 total)
Document pipeline
OCR + extraction + summarisation for regional scripts
Private-cloud option
Customer-VPC deployment template
Analytics dashboard
CSAT, AHT, resolution, intent drift, language mix
Acceptance
1 internal pilot live · 10K+ real conversations served · compliance posture review complete
Explicitly out of scope for v1 (6 months)
18 · Phased Roadmap
Two phases.
Six months. One production cutover.
Phase 1 proves the engine. Phase 2 makes it production-grade and onboards the first internal pilot. Each milestone is a checkpoint for steering review.
M1
Phase 1
Team assembled · stack chosen · core repo live
M2
Phase 1
ASR + TTS + NLU pipeline · 3 languages working
M3
Phase 1
Phase 1 demo · 5 languages · 1 vertical adapter
M4
Phase 2
Multi-tenant + agentic workflows · pilot tenant onboarded
M5
Phase 2
Live agent assist · analytics dashboard · 10 languages
M6
Phase 2
Production v1 cutover · compliance posture reviewed
19 · The Horizon
Cognitive Intelligence Inside
Made for the world.
From Nepal to India, Sri Lanka to Bangladesh, Malaysia to Indonesia, the Gulf to North Africa — wherever language is the barrier between people and business, SABAI is the bridge.
Phase 01
South Asia
Nepal · India · Sri Lanka · Bangladesh
Phase 02
ASEAN
Malaysia · Indonesia · Singapore · Thailand
Phase 03
MENA
UAE · Saudi · Egypt · Qatar
SABAI को लागि.
SAB के लिए AI.
For everyone.
Untuk semua.
سب کے لیے.