TalkifAI

Context

TalkifAI was a Voice AI platform I founded and built end to end. It was used by SIEHS, the Sindh government's emergency health service, and HealthCloud. The platform operated during 2025–2026 and has since closed.

The engineering work extended beyond the conversation itself: an agent needed telephony, a usable product interface, and enough operational visibility to understand what happened during a call.

My responsibility

I built the agent orchestration, telephony integration, dashboard, documentation, and billing. This connected the real-time agent with the product people used to configure, operate, and review it.

That scope required both implementation and product judgment: the call runtime and the surrounding product had to work together.

How the system fit together

• Agent orchestration and telephony handled the live voice interaction.
• RAG memory was part of the system's context layer.
• The dashboard exposed per-call recordings, transcripts, and analytics, including call duration.
• Documentation and billing supported the product around the agent.
• Instrumentation covered the API, telephony, database, recording and egress, batch jobs, and RAG memory.

These were parts of one product, with operational visibility across the conversation and the services supporting it.

Reliability and verification

I implemented automated end-to-end call testing and public component-level status reporting. The reported uptime was 99.9%.

That figure is a historical reported metric, not a current service guarantee. It should be read alongside the component-level reporting rather than treated as a promise that every individual call would succeed.

Per-call recordings, transcripts, and analytics provided another view of behavior: what a conversation contained, how long it lasted, and what a user could review after it ended.

Capacity and outcome

The system was provisioned for 20 simultaneous calls, approximately 400 calls per hour, and supported 90+ languages. Provisioned capacity is distinct from measured peak traffic; these figures do not claim sustained customer usage at that rate.

SIEHS and HealthCloud used the platform. TalkifAI closed in 2026, but the work demonstrates the engineering scope of a complete Voice AI product: real-time runtime, business integrations, user interface, testing, and operations.

What this brings to a team

I can take responsibility for an agent beyond its prompt: how it connects to telephony, how users operate it, and how the team observes and tests it. That makes the work relevant to an engineering role owning a production voice feature or a larger agent platform.