Engineering, Voice AI

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.

Check out our principles https://principles.telli.com/ to see how we work.

ABOUT THE ROLE

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
• realtime audio
• 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.

Multi-channel agents

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

Analytics & Dashboard

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.

OUR TECH STACK

• 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
• Infrastructure: AWS, GCP, Azure, Cloudflare, Render, Cerebrium, Modal

WHAT MAY NOT MAKE YOU A GOOD FIT

• you prefer a remote-first environment
• you want to be told exactly what to do
• you mostly care about the tech, not business outcomes
• you want to work 9-5

PERKS

• very competitive pay + equity package
• access to any (AI) tools
• large token budget
• free dinners if you stay late
• urban sports
• relocation support if you don’t live in Berlin

We show an excerpt and a link, and the full text when the source allows it. Brands and logos shown belong to their owners.

Ancient Origins · 11 days ago

Posted Nov 01, 2026

Posted Nov 01, 2026

Source: Statistisches Bundesamt (Destatis), table; dl-de/by-2-0; StarIn's calculation