Software Engineer, Payments Infrastructure

What you'll do

The agent must collect and use a card number but can never see it. This work sits across four hard technical domains: payments, security, real-time voice, and LLMs. Voice is the core challenge. A caller speaks or types a card number in a live, low-latency audio stream, and the digits must never reach the model, while the model stays on the call.

Sierra owns every system that touches card data, and one small team owns it end to end across three workstreams: Platform, Product, and Safety & Integrity. Customers keep their own telephony and payment gateways. Payments is live in production for some of the largest consumer enterprises, on a platform that handles hundreds of millions of calls a year. The platform is certified to PCI DSS Level 1.

• Build real-time card capture for voice: intercept keypad and spoken digits in a live audio stream, upstream of the agent runtime in a network-segmented environment, while the LLM stays live on the call. Make it work across every telephony provider and protocol Sierra supports.
• Build and extend the tokenization platform: agents receive and send short-lived tokens, and the platform swaps a token for the real value only when it calls the customer's payment gateway.
• Own the payments egress control plane: every destination for card data is approved, versioned with the agent release, and enforced by infrastructure below the application, not by convention.
• Make payment actions correct: an at-most-once action primitive in the Agent SDK, validated amounts and items, and monitoring that compares what the agent said about a payment with what really happened.
• Keep card data out of places it should not be: deterministic detection in the agent's hot path that catches what general-purpose models miss. No LLM, including any we build ourselves, ever sees a plaintext card number.
• Make payments self-serve: one payments API across Agent Studio, the Agent SDK, and Ghostwriter, Sierra's agent-building agent; gateway middleware that developers write against tokens; and support for single-tenant and customer-cloud deployments. Any team can add payments to an agent without a payments engineer in the loop.
• Grow the platform beyond cards: extend tokenization into a privacy vault for all PII.

What you'll bring

• 5+ years building backend, platform, or infrastructure systems, with a track record of shipping and operating high-reliability services in production.
• A deep security instinct. You design as though every component will eventually be compromised, and you reason naturally about blast radius, least privilege, and making bad states impossible rather than reacting to them.
• Rigor about correctness under failure: retries, idempotency, at-most-once behavior, and clear handling of timeouts and unclear results from external systems.
• Fluency in a systems language (we use Go) and comfort owning critical systems end to end, from design through production and on-call.
• Range and ownership. You like a small team with broad scope, you build clean abstractions over messy external systems, and you're never satisfied solving a problem once per customer.
• Clear communication. You can explain a design or a risk to other engineering teams and to non-engineers.
• Degree in Computer Science or a related field, or equivalent professional experience.

You don't need payments experience. The engineers who built this platform came from backend, infrastructure, and security work, not payments. We care more about how you build correct, secure systems than about the word "payments" on your resume.

Even better…

• Real-time or streaming systems: voice, telephony, media, or low-latency pipelines.
• Applied cryptography and key management, network isolation, egress control, or cloud security.
• Platforms other engineering teams build on, such as APIs, SDKs, and self-serve developer services.
• Deployments into customer-owned clouds or single-tenant environments.
• Customer-facing technical work, such as architecture reviews with customers.
• Setting technical direction on ambiguous, high-stakes problems.
• Payments, fintech, or another sensitive-data domain such as healthcare or identity.