Principal AI / LLM Engineer – Agentic Orchestration

Hiring for: A US based well funded startup building low-latency, agentic voice AI systems at production scale.

Role: Principal AI / LLM Engineer – Agentic Orchestration

Experience: 8yrs+

Location: Bengaluru

Type: Hybrid - 3 days Work From Office

Notice: Immediate preferred, no longer than 30 days

Salary: Based on fitment (75L range + massive ESOPs)

Role:

Design and own the orchestration layer (control plane) powering production AI agent systems — building multi-step workflows that coordinate models, tools, and services with deterministic execution and controlled failure.

This role sits at the intersection of AI systems and workflow orchestration, where LLM-driven agents interact with APIs, data systems, and internal services through structured execution pipelines.

Here's what you're EXTREMELY good at:

• Shipping production AI agent systems (not prototypes or experiments) that orchestrate models, tools, and services
• Designing multi-step agent workflows with deterministic execution and clear state transitions
• Building orchestration layers that coordinate LLM reasoning, tool execution, and system state using patterns such as state machines, schedulers, or operator-style reconciliation loops
• Modeling state machines, dependency chains, and workflow DAGs
• Designing guardrails, invariants, and safety boundaries for AI agent behavior
• Handling retries, idempotency, compensating actions, and partial failures in complex workflows
• Building systems that remain reliable when agents interact with external tools and APIs
• Owning systems end-to-end: design → production rollout → incident recovery
• Working with event-driven architectures, workflow engines, distributed schedulers, or orchestration platforms (e.g. Temporal, Cadence, LangGraph, LangChain, LlamaIndex, Airflow, Step Functions, Argo or similar)

Here's how your peers and manager describe you:

• Thinks in states, transitions, and constraints, not just function calls
• Instinctively asks “How does this break?” before discussing how it works
• Decomposes complex AI systems into clear orchestration layers and execution components
• Designs abstractions that remain stable under scale, concurrency, and partial failure
• Understands that agent behavior must be deterministic and observable in production
• Moves fast without sacrificing structural integrity
• Adapts your model when challenged instead of defending weak assumptions

This is NOT who you are:

• Someone experimenting with LLM APIs without having shipped production AI systems
• A prompt engineer focused only on model tuning or chaining APIs
• A research-only AI/ML profile without production system ownership
• A generic backend engineer whose work stops at CRUD microservices
• A people manager stepping away from hands-on system design