A complete application in a minute
Tailored resume and cover letter, ready to send.
tl;dv is hiring a Senior Software Engineer, Voice AI to advance speaker recognition and transcription in a fast-growing startup. You will ship robust models, measure performance, and improve accuracy across multilingual meetings. Your work impacts millions of meetings and user trust.
You will own end-to-end speaker systems, manage production deployments, and collaborate with product and security teams in a fully remote, async-first environment based out of Brazil.
Context tl;dv helps revenue, customer, product and leadership teams turn meetings into structured intelligence. We record, transcribe, summarise and analyze meetings, so teams understand what's happening across sales demos, feedback calls, onboarding sessions and internal discussions. We're growing fast: revenue is up 5x and usage 10x over the last 12 months. Everything we do starts with two questions: what was said, and who said it.
We're hiring a Senior Software Engineer, Voice AI (Speaker Recognition & Transcription) to join our Transcription team. You'll focus on two things: transcription (what was said) and speaker recognition (who said it). This is one of the most important roles at tl;dv. Every summary, insight, CRM update and AI answer we produce starts from the transcript and the speaker labels. If a word is wrong or a sentence is attributed to the wrong person, everything built on top of it is wrong too. Getting this right is what earns our customers' trust with their most important conversations, and it's one of our top product priorities. Your work will be felt by every user, in every meeting. Meetings are messy: noisy audio, short turns, people talking over each other, many languages. Your job is to make our transcripts and speaker attribution more accurate, faster and more scalable, and to prove it with numbers. This is a hands-on role with a lot of autonomy. You'll own problems end-to-end, from reading the latest specs to shipping in production and measuring the results. You'll report to the CTO and work closely with Product, Infra, QA and Support.
Own our speaker recognition and diarization systems end-to-end, from model choice to production. Make them robust to real meetings: noisy audio, short turns, overlapping speech, several languages. Evaluate and bring in better speaker models (open-source or in-house) when they beat what we have.
Improve transcript quality across our speech-to-text models and meeting platforms: accuracy, speaker attribution, punctuation, custom vocabulary and multilingual support. Help decide which model runs where, balancing quality, latency and cost.
Build and own our evaluation framework: datasets and metrics for transcription and speaker recognition. Every model change should be measured, not guessed.
Run these models in production at scale for millions of meetings, with good observability, sensible cost and no surprises.
Voice data is sensitive data. You'll build with consent, data isolation and deletion in mind from day one, working closely with our security and privacy team.
You've scaled speech and voice systems in a fast-growing startup and want to do it again. You're super hands-on and highly motivated, and you don't need anyone to tell you what to do next. You see a problem, measure it, fix it and ship it.
Experience with audio processing: noise reduction, voice activity detection, speech enhancement, overlapping speech.
Multilingual or low-resource speech experience.
Experience handling voice or other sensitive data under GDPR (consent, retention, deletion).
Event-driven architectures (message brokers, queues).
Hands-on experience with speech toolkits such as pyannote, SpeechBrain or NVIDIA NeMo.
Publications, open-source contributions or results in speech/speaker benchmarks (e.g., VoxCeleb, NIST SRE, DIHARD).
Type: Full-time
Engagement: We're open to both a contractor setup and employment through an Employer of Record (EOR). We'll figure out together what works best for you.
Location: Fully remote, based in Brazil
Ways of working: Async-first. We default to written communication (docs, RFCs, recorded walkthroughs) and keep meetings for the moments that need them.
Reports to: CTO