דרושים » תוכנה » AI Architect, Agentic platform

Required AI Architect, Agentic platform
The Role:
You own the end-to-end architecture of the agentic platform: how speech, avatar generation, and the agentic reasoning layer combine into one coherent system that holds up in production, across many tenants, under a real-time latency budget.
This is the widest technical seat in the group. A single user turn crosses speech recognition, retrieval, planning and tool calling, speech synthesis, and avatar rendering, and the experience is only as good as the seams between them. The scope here is the whole path, including the parts we did not build ourselves: externally sourced capabilities, third-party model providers, and customer-owned systems reached through the gateway.
Some of the decisions here shape the platform for a long time: how the speech pipeline is composed, which real-time orchestration and transport framework the platform standardizes on, how new capabilities are absorbed into the target architecture, and where a shared component belongs versus a per-customer one. You define the criteria these decisions are judged against and work closely with the research team that benchmarks the options, so the call rests on evidence. You write the reasoning down, and you stay close enough to the code to know when reality disagrees with the design.
You are hands-on. You prototype to de-risk, you read the traces yourself, and you stay as close to the numbers as to the diagram.
What Youll Do:
Own the end-to-end system architecture across the real-time conversational path and the offline video path, defining the interfaces and boundaries between the speech, avatar, and agentic layers.
Drive the decisions that set platform direction, including speech pipeline composition, real-time orchestration and transport frameworks, model routing, and build-versus-adopt calls. You frame the criteria and the trade-offs, partner with the research team on the benchmarks that test them, and own the resulting call.
Lead the architecture for absorbing new capabilities, whether built, adopted, or acquired, and define the converged target architecture.
Architect for latency and cost as first-class constraints, owning the end-to-end latency budget, the cost-per-outcome model, and how the system degrades when a component stalls or fails.
Design the long-term memory architecture and the privacy model behind it: consent, retention, erasure, and residency.
Define cross-platform interfaces and engineering standards that let the foundation team and the FDE pods move independently, including the registry contract, hooks, and the governed path for customer-owned components.
Set the technical governance model for how capabilities are evaluated, promoted, versioned, and retired, with evaluation and tracing designed in rather than added afterward.
Prototype and de-risk future capabilities through focused technical spikes, turning results into clear recommendations.
Partner across the group and the company, from the research, platform, and FDE teams to product, and represent the architecture credibly to executives and customers.