Engineering, Voice AI Berlin

telli; Berlin; Vor Ort; EUR 70.000 - 110.000; Vollzeit

Benefits dieser Stelle

Relocation support if you don’t livein

Zusammenfassung

telli is seeking an engineer to build fast, reliable Voice AI and backend systems at Berlin-scale. You will own end-to-end projects from idea to production, shaping core product decisions and driving real customer outcomes.

You should ship code, solve distributed systems challenges, and collaborate across frontend, backend, and data teams. Proficiency in Python/TS, FastAPI, PostgreSQL, Redis, and SQLite is valued.

Qualifikationen

• Built and owned distributed systems in production.
• Strong software fundamentals and ability to learn unfamiliar domains quickly.
• Eager to act, bias toward taking initiative and getting hands dirty.

Aufgaben

• Develop Voice AI and backend components that are fast, reliable, and high-quality.
• Build an distributed call scheduler to manage load and avoid conflicts across customers.
• Develop agent memory to persist and access context across conversations.
• Enable Charlie to configure, debug, and improve agents from the app, Slack, and ChatGPT.
• Create workflows to connect agents to CRMs, calendars, emails, and customer systems.

Kenntnisse

Distributed systems

Python

TypeScript

FastAPI

Node.js

PostgreSQL

Redis

SQLite

OpenTelemetry

LiveKit

React

AI basics

Tools

LiveKit

SQLite

PostgreSQL

Redis

Temporal

OpenTelemetry

Honeycomb

Better Stack

Jobbeschreibung

Every 10 minutes, 18 million conversations happen between businesses and consumers around the world.

For decades, the consumer side of those conversations sucked: long queues, robotic menus, repeated explanations, and advice that is often unhelpful. Most people today still cannot simply say what their problem is in natural language when communicating with businesses.

AI changes that. For the first time, companies can have natural conversations with customers at massive scale. Every B2C company - from energy providers to telcos to insurers - will use AI agents to talk to their customers. The number of conversations businesses will handle is going to explode in the coming years.

At telli, we are solving the challenges that come with that shift. Today, leading B2C companies like Sky are already using telli to deploy thousands of voice agents to provide their customers with a new experience. But building a first voice agent is easy, getting it to drive real outcomes for customers and businesses is the hard part. telli helps companies build, deploy, and improve consumer-facing AI voice agents at scale. This will become one of the most important software categories of the next decade, and we intend to win it.

We are a small, AI-pilled team that likes to solve hard problems. Be it engineering, product, or GTM - we build, experiment, and move fast, while heavily leveraging the capabilities of AI models. All of us like to challenge ourselves and take real ownership over what we do, all while still staying humble and keeping a low-ego culture.

This role is for engineers who care about velocity and solving real customer problems. You won't just ship code; you will own projects end-to-end, make core product and technical decisions, and shape the trajectory of this company.

Our philosophy is to push the frontier models to the limits of what is possible - and to achieve that, we need your help.

We are a small team working across:

harness engineering

applied AI products

distributed systems

what you will work on

Voice AI

Voice AI is one of the hardest technical challenges we have right now. We need to ensure that the voice AI we deliver is fast, reliable, and high-quality.

Topics that we are currently working on:

No-infra RAG using SQLite artifacts downloaded for each call.

Racing LLMs to use the fastest one at any given moment.

Shadow-test new transcribers on real calls before switching production traffic.

Make Voice agents respond faster without reducing quality.

Make every bad call easy to understand and debug.

Currently, our agents are primarily voice agents, but we will extend their capabilities to be able to handle any channel.

Build an agent that is able to call someone, follow up on WhatsApp, wait three days for a reply, send an email, and continue exactly where it stopped.

Let agents choose the right channel instead of forcing every interaction into a phone call.

Share memory, tools, and state across realtime and asynchronous conversations.

Call scheduler

We have to develop a distributed call scheduler that will manage load across our voice orchestration. It has to handle many race conditions and distribute the load evenly across customers while applying additional rules.

Start large call batches without letting one customer consume all available capacity.

Prevent duplicate calls even when workers crash, restart, or race each other.

Handle time zones, calling windows, retries, provider limits, and changing capacity.

Keep the system moving when individual calls, workers, or providers fail.

Agent memory

Our agents have multiple interactions with customers, so we need to ensure they can reason throughout the whole flow.

Topics we have to develop:

What information should be persisted in memory?

How should the voice agent access memory?

Should agents share memory?

Charlie

Charlie is our AI coworker for building, configuring, debugging, and improving agents.

Enable Charlie to use every capability in our product without building one-off tools.

Make Charlie available through the app, Slack, Teams, ChatGPT, and Claude.

Make every action token-efficient, traceable, and easy to understand.

Enable Charlie to use customer-context to make effective proactive suggestion on improving agents

We want Charlie to replace dashboards full of charts with direct, useful answers.

Enable Charlie to turn plain-language goals into business KPIs.

Give Charlie the context to explain metric changes and trace them to exact calls.

Enable Charlie to surface trends, regressions, and opportunities automatically.

Workflows

Voice agents need to do real work, not just have conversations.

Enable Charlie to connect agents to CRMs, calendars, WhatsApp, email, and customer systems.

Give Charlie the ability to build workflows that wait, retry, and resume.

Enable Charlie to inspect failures, explain their cause, and suggest a fix.

what we look for

We believe that today, engineers must not only know how to ship code but also ensure they work on the right problems. These are the core skills we look for in the interview process:

You have built and owned distributed systems in production.

You have strong software fundamentals and learn unfamiliar domains quickly.

You are eager to act, biased towards action, and want to get your hands dirty.

You want to figure things out on your own rather than waiting for people to tell you what to do.

You don’t take what people say as given; you want to understand the problems yourself and challenge the status quo.

You are excited about agentic coding, but can do it with awareness ensuring quality

You are a team player: You optimize for the engineering organization’s output, not your individual output.

Voice and backend: Python, uv, FastAPI, LiveKit, TypeScript, Bun, Node.js, Fastify

Frontend: React, Vite, TanStack

Data and workflows: PostgreSQL, Redis, Temporal, SQLite

Observability: OpenTelemetry, Honeycomb, Better Stack

what may not make you a good fit

you want to be told exactly what to do

you mostly care about the tech, not business outcomes

you want to work 9-5

very competitive pay + equity package

access to any (AI) tools

urban sports

relocation support if you don’t live in Berlin