Generative AI Engineer

Company Description

Booked AI is an AI-native corporate travel platform. Employees book flights, hotels, and ground transport through natural conversation across web chat, WhatsApp, Slack, Outlook, and voice. Companies receive policy-aware guidance, centralized billing, and administrative controls through the agent and a dedicated dashboard.

The agent is the primary product interface. It searches live inventory, coordinates multi-leg itineraries, manages checkout and payment, and handles post-booking servicing including cancellations and refunds. Reliability is achieved through deliberate system design: lean prompts, a consolidated tool surface, and authoritative data retrieval; not through layered orchestration

Role Description

We are seeking an AI Engineer to own and extend our agent runtime: model reasoning, tool execution, multi-channel streaming, and quality measurement at scale.

This is a backend-focused role (approximately 80% Python across agent runtime, MCP tooling, and channel integration). You will collaborate with frontend engineers on chat streaming and structured UI components, but the primary remit is agent systems engineering.

Responsibilities

• Extend the core agent runtime: how sessions are held, how turns stream, how models are selected, and how state survives across conversations and channels
• Evolve the MCP tool surface via structured discriminators rather than proliferating primitive schemas
• Implement reasoning-first structured intent contracts to ensure the agent resolves bookings, policies, and inventory from tools and database records
• Expand scenario-based evaluations, LLM-as-judge pipelines, and regression coverage for booking, checkout, and policy workflows
• Operate tracing, skill telemetry, and production diagnostics for tool-call consistency and intent enforcement
• Implement travel-domain logic including package drafts, multi-city routing, offer revalidation, payment handles, and cancellation/refund flows via content providers and platform specific APIs.

Preferred Qualifications:

• Experience designing tool surfaces and execution boundaries for LLM agents in production, not just wiring API calls
• Track record shipping multi-surface systems where one backend serves several client protocols without forking business logic
• Comfort operating in high-stakes transactional domains where stale state, duplicate execution, or wrong identity binding has real cost
• Realtime systems experience: bidirectional streaming, session affinity, interruption handling, and latency-sensitive turn loops
• Built or maintained eval infrastructure that catches regressions before users do — scenario runners, trace analysis, automated judges, failure taxonomy
• Strong production Python: async concurrency, schema design, idempotency, observability, and knowing when not to add another abstraction layer
• Clear judgment on what belongs in model reasoning versus what must be enforced by deterministic execution boundaries

Engineering Principles

1. Correctness before convenience: Engineering decisions are judged by whether the system behaves correctly under real conditions, not by whether the first implementation was fast to ship.
2. Explicit over implicit: Interfaces, ownership, and failure modes are defined clearly. Behaviour should not depend on undocumented assumptions, hidden coupling, or tribal knowledge.
3. Simplicity as a discipline: Complexity is added only when it is justified. The preferred solution is the smallest one that remains correct, observable, and maintainable.
4. Ownership through production: Engineers are accountable for outcomes after release: reliability, operability, and the cost of the systems they introduce. Building and running are not separate responsibilities.

Location and Capacity:

• Location: Benaluru, Karnataka, India
• Employment: Full-time, on-site