You'll join the team that owns Telnyx's phone-number inventory and the entire ordering pipeline behind it. The stack is Python, the domain is global telephony, and the roadmap is near-total automation: if a step in the pipeline is done by a human today, we're building the system that takes that step away.
What You'll Own
The ordering pipeline** — async Python services that turn a search-and-order API call into a provisioned, billable phone number: job queues, retries, rollbacks on partial failure, SLA clocks, and backorder (advanced order) fulfillment where lead-time promises have contractual weight. Tens of thousands of orders a week run through this today.
Vendor & carrier integrations** — dozens of them, from US Tier-1 wholesale carriers to international carriers and regulated-market vendors. Each integration is its own Snowflake — idiosyncratic APIs, failure modes, and regulatory constraints. You'll make them behave like one system.
The AI order-review layer** — a production LLM service that reads orders and customer-submitted regulatory documents and auto-approves by country, number type, and carrier, governed by allowlists, score thresholds, and full audit logs. You'll own expanding its coverage safely: evals, guardrails, regression suites, and clean handoffs to humans when the machine isn't sure.
Regulatory automation** — per-country identity/documentation checks ("requirement groups"), external document-verification vendors, expiry and re-verification flows — automated end-to-end in 45+ countries we hold carrier licences in.
The data layer** — a large Postgres-based warehouse and Metabase analytics serving stock, SLA, and order-lifecycle dashboards used daily by ops and leadership, across a 30+ microservice estate; alerting on stock risk, vendor SLA breaches, and pipeline health.
Tech stack (what you'll actually touch)
Python 3 (asyncio, FastAPI/aiohttp, SQLAlchemy, Alembic, pydantic), pytest, mypy/ruff
LLM tooling: model APIs, structured output, agent orchestration, prompt/scoring configuration, offline eval harnesses
PostgreSQL at scale: schema design, migrations on live systems, query tuning, warehouse SQL
RabbitMQ/Kafka, Consul, Kubernetes, Jenkins, Docker/OCI, GitHub Actions
Metabase, and TypeScript/Next.js for internal dashboards (not required — the backend is yours)
What we're looking for
5+ years of backend engineering, with strong production async Python.
You've owned systems that talk to many external services and lived with the async-job, retry, rollback, and eventual-consistency problems that creates.
You can make an LLM trustworthy in a real flow: validation, scoring thresholds, evals, audit, and human-in-the-escalation. You know where AI is reliable and where it's theater.
Postgres beyond ORM basics — you can read a plan, write a migration on a live system, and know why SKIP LOCKED matters.
You're a pragmatist about process automation — a workflow isn't done when the ticket closes, it's done when the human step is gone and the numbers prove it.
Strong written communication across a distributed team (EMEA + Americas).
Bonus points for
Telecom/CPaaS background: DID inventory, ordering, porting (NPAC), carrier provisioning, regulatory/KYC familiarity.
Experience with document/KYC verification vendors (Onfido, Sumsub, Jumio) or OCR pipelines.
Building with agents beyond work: MCP servers, tool-calling frameworks, open-source agent runtimes.
Warehouse/analytics engineering; internal-dashboard work in TypeScript.
What this role is not
Not a pure CRUD role. Every row in this system is a real phone number somewhere — on a carrier, under a regulator, promised to a customer, watched by an SLA.
Not an ML research role. You'll build applied AI that approves real transactions — no notebooks in production, no fine-tuning required.
