Senior AI Software Engineer

At HeyDiga we build conversational AI agents that answer the phone and WhatsApp for service businesses in Spain, Italy and France. Our product is called SONIA, and we don’t sell it as software: she’s a virtual employee who picks up calls, manages appointments, answers questions and never leaves a call unanswered. We work with hair salons, beauty centres, car repair shops, restaurants and clinics.

Every day, SONIA holds thousands of real conversations with real customers. Behind every call there’s a full machine learning pipeline: automatic speech recognition (ASR), an LLM that reasons and makes decisions, and text-to-speech (TTS).

The role

Your job is to make SONIA not just talk and reason well, but do it fast and reliably across thousands of real calls and WhatsApp conversations every month. No demos: what you build goes to production and a real customer hears it.

• Core platform: design, build and operate the APIs and services in TypeScript (Node.js / Bun) that power real-time voice conversations - scalable, reliable and secure, because every failure is a call nobody answers.
• AI pipelines in production: integrate and orchestrate LLMs, STT and TTS (OpenAI, Whisper, ElevenLabs and our own models) - streaming, caching, fallbacks, prompt versioning and voice edge cases such as retries on failed transcriptions, fallback voices and latency budgets.
• Agentic workflows: design agent and sub-agent workflows, chain outputs, manage context and tools, and apply Spec-Driven Development and/or Research-Plan-Implement (RPI) in your daily work.
• Product, end to end: build with React and Tailwind the surfaces our customers and internal team use - dashboards, analytics and agent configuration tooling.
• Infra & observability: cloud (AWS/GCP), deployment pipelines, monitoring, token-usage tracking and logging for real-time voice processing.
• Quality and technical judgement: clean architecture, DDD, hexagonal design, auth/authz, rate limiting and tests where they really matter. Mentor teammates and help define how we work.

You will work directly with the founders, ship to production daily and have a real say in how the product, the platform and the team evolve.

Who we’re looking for

• 5+ years of production experience with modern web stacks.
• You use AI coding agents (Claude Code, Cursor, Codex or similar) as a core part of your daily workflow - not as an occasional helper - and you understand how they work under the hood: tokens, context windows, sub-agents, tools and orchestration.
• Experience designing agentic or multi-agent workflows.
• Solid grasp of relational databases, SQL and data modelling; comfortable with Docker and containerized environments.
• You know and apply Clean Architecture, DDD and hexagonal architecture, and can set up a sensible testing strategy (Jest, Cypress or similar).
• Autonomous, pragmatic and product-minded: comfortable with ambiguity and able to ship fast without sacrificing quality.
• Fluent Spanish and English, written and spoken.

Nice to have

• Production experience integrating LLMs, STT and TTS (OpenAI, Whisper, ElevenLabs, Deepgram or similar).
• Real-time / streaming systems, low-latency pipelines, WebSockets or WebRTC.
• AI orchestration frameworks (mastra, LangChain, LangGraph or similar).
• Polyglot backend: Python, Go, Ruby or Java.
• System design and DevOps: microservices, API security, CI/CD, observability (Datadog, Sentry, PostHog).
• Previous experience as a founding or early engineer in a venture-backed startup.

Our stack

• Runtime: Node.js, Bun
• Frontend: React, Tailwind CSS
• Data: relational databases (SQL)
• Infra & DevOps: Docker, AWS / GCP, CI/CD
• AI layer: multi-LLM orchestration, STT/TTS (OpenAI, Whisper, ElevenLabs, custom models), prompt versioning, evals, fallbacks

What we offer

• Meaningful equity in an early-stage, venture-backed company.
• Top-tier hardware of your choice.
• Work side by side with senior founders who have built and scaled top Spanish tech companies.
• Real ownership over technical decisions, product surface and engineering culture.
• An AI-native engineering culture: we don’t just build AI, we use it intensively to build faster and better.

• 5+ years of production experience with modern web stacks.
• You use AI coding agents (Claude Code, Cursor, Codex or similar) as a core part of your daily workflow - not as an occasional helper - and you understand how they work under the hood: tokens, context windows, sub-agents, tools and orchestration.
• Experience designing agentic or multi-agent workflows.
• Solid grasp of relational databases, SQL and data modelling; comfortable with Docker and containerized environments.
• You know and apply Clean Architecture, DDD and hexagonal architecture, and can set up a sensible testing strategy (Jest, Cypress or similar).
• Autonomous, pragmatic and product-minded: comfortable with ambiguity and able to ship fast without sacrificing quality.
• Fluent Spanish and English, written and spoken.

Nice to have

• Production experience integrating LLMs, STT and TTS (OpenAI, Whisper, ElevenLabs, Deepgram or similar).
• Real-time / streaming systems, low-latency pipelines, WebSockets or WebRTC.
• AI orchestration frameworks (mastra, LangChain, LangGraph or similar).
• Polyglot backend: Python, Go, Ruby or Java.
• System design and DevOps: microservices, API security, CI/CD, observability (Datadog, Sentry, PostHog).
• Previous experience as a founding or early engineer in a venture-backed startup.

Our stack

• Runtime: Node.js, Bun
• Frontend: React, Tailwind CSS
• Data: relational databases (SQL)
• Infra & DevOps: Docker, AWS / GCP, CI/CD
• AI layer: multi-LLM orchestration, STT/TTS (OpenAI, Whisper, ElevenLabs, custom models), prompt versioning, evals, fallbacks

• Work side by side with senior founders who have built and scaled top Spanish tech companies.
• Real ownership over technical decisions, product surface and engineering culture.
• An AI-native engineering culture: we don’t just build AI, we use it intensively to build faster and better.

At HeyDiga we build conversational AI agents that answer the phone and WhatsApp for service businesses in Spain, Italy and France. Our product is called SONIA, and we don’t sell it as software: she’s a virtual employee who picks up calls, manages appointments, answers questions and never leaves a call unanswered. We work with hair salons, beauty centres, car repair shops, restaurants and clinics.

Every day, SONIA holds thousands of real conversations with real customers. Behind every call there’s a full machine learning pipeline: automatic speech recognition (ASR), an LLM that reasons and makes decisions, and text-to-speech (TTS).