KINDIكندي

Engagement Blueprints

The shape of the work, with the names left out.

We don’t trade on client names — engagements are confidential by default. What we can publish is the blueprint: the scope, the sequence, the team, and the numbers we commit to move.

01Government

Arabic-first citizen assistant for a service ministry

A ministry serving millions of citizens, with a contact center absorbing routine inquiries that could resolve in seconds, and a mandate to serve citizens in their own language — dialect included, not just fusha.

Team:
4 engineers · 1 lead · ministry SMEs embedded
Timeline:
12 weeks to assisted launch, 6 months to full rollout

What gets built

  • Dialect-aware assistant across web and WhatsApp, grounded in the ministry’s actual policy corpus
  • Escalation design: uncertain or sensitive cases route to human agents with full context attached
  • Arabic evaluation suite covering MSA and Gulf dialect variants, gating every release
  • Sovereign deployment with complete audit trails for oversight bodies

The numbers we sign up to move

  • Containment rate on routine inquiries, measured against the call baseline
  • Arabic answer accuracy vs. the eval suite, published internally per release
  • Median citizen wait time for the cases that still need a human
02Financial Services

Agentic back-office for a retail bank’s exceptions desk

A bank whose operations team clears thousands of payment exceptions, disputes, and reconciliation breaks monthly — experienced staff spending most of their day gathering context that systems already hold.

Team:
3 engineers · 1 lead · operations and compliance embedded
Timeline:
10 weeks to shadow mode, 16 weeks to tiered autonomy

What gets built

  • Agent workflows that assemble case context, draft resolutions, and file the audit record
  • Risk-tiered autonomy: low-value clears run straight through, high-value queue for approval
  • Evaluation harness on historical cases before the system touches a live one
  • Full action logging designed with compliance in the room from week one

The numbers we sign up to move

  • Cost and cycle time per case vs. the pre-agent baseline
  • Straight-through rate at each autonomy tier, expanded only on eval evidence
  • Zero unaudited actions — the number that must stay at zero
03Enterprise

LLM platform consolidation for a multi-division group

A group where five divisions built five AI stacks with five vendors — spend growing faster than value, security reviewing everything twice, and no one able to say what a use case costs.

Team:
5 engineers · 1 lead · group IT and security embedded
Timeline:
3 weeks assessment, 10 weeks core platform, 6 months full migration

What gets built

  • One governed gateway: routing, fallbacks, caching, and per-division cost attribution
  • Shared retrieval and fine-tuning pipelines inside the group’s residency boundary
  • Migration of existing workloads without breaking the teams that built them
  • A platform playbook so the next use case onboards in days, not procurement cycles

The numbers we sign up to move

  • Cost per use case, visible for the first time and trending down
  • Time-to-production for a new use case, in weeks not quarters
  • Share of model traffic through the governed gateway, driven to 100%

These blueprints describe how Kindi structures engagements of each shape — scope, sequencing, staffing, and success metrics. Baselines and results are always measured per client and agreed before build; our approach explains how.

Which blueprint is closest to your problem?

Tell us, and we’ll adapt it to your systems, your regulator, and your baseline — in writing, before you commit to anything.

Book a strategy call or write to salam@kindihq.com