AI/ML Platform Engineer

About Halcyon Labs Founded in 2016 in Kraków, Poland , Halcyon Labs is a technological company working with major European banking , insurance , and telco enterprises . The company was set up with an ambitious goal of providing the best customer service in the world . Lekta stands out from the crowd of conversational AI companies because we combine technological expertise with domain-specific knowledge from contact center industry veterans . Through the use of our proprietary conversational neurosymbolic engine and backend integrations , our systems are intent- and context-aware , handling over 1.5 million conversations every month . Today, with offices in Poland and Spain , Lekta is rapidly expanding its business division to scale operations internationally . About the role Lekta is hiring an AI/ML Platform Engineer to own the full AI/ML stack behind our conversational agents — language models , ASR and TTS , retrieval and grounding , and the eval discipline that ties them together. The work is shaped by real constraints: ultra-low latency on voice channels , where every 100 ms is felt at telco and mobile scale; the architectural calls our neurosymbolic design demands, where you decide what belongs in deterministic logic and what belongs to the language model ; and the bar that regulated domains — banking, insurance, telco — set for auditability and predictable behaviour . You will also be the team's eye on the frontier — model releases , new techniques , and research worth integrating . We are building the next generation of conversational AI — a platform where the creator's experience is itself AI-native . Think of what Lovable did for app-building, applied to conversational agents: a visual IDE , a copilot that helps the user shape, refine and evolve the agent, and a runtime engineered to deploy what gets built into enterprise production . What makes the engineering genuinely hard is the combination our agents have to deliver simultaneously: fluent (natural language, real conversation), fast (sub-second on live voice channels), and reliable (predictable behaviour under real customer pressure, every time). Most stacks force a choice between two of those at the cost of the third. We don't , and that constraint shapes almost every engineering problem we work on. AI is a power tool , and we treat it like one. Lekta engineers are encouraged — and equipped — to use AI deeply across their work: Claude Code , model APIs , copilot tooling . We invest in the accounts, tooling, and time it takes to get fluent with them. We don't measure people on lines committed ; we measure on what gets delivered and how well it holds up in production . We are not an agentic-workflow company . AI generates; the engineer directs and reviews. Every change going out under your name has been read, understood, and signed off by you . Responsibilities Own the language model layer — provider integration , model selection , prompting , evaluation , and cost-quality trade-offs at scale Own the speech stack — ASR and TTS pipelines , latency budgets , multilingual quality Own the techniques that take voice to ultra-low latency — incremental dialogue management , end-of-turn detection , and the engineering that closes the gap to natural conversation Lead retrieval and grounding — embeddings , RAG , recall-precision trade-offs at production scale Make the architectural calls between deterministic logic and language model invocation , and the composition between them Track the frontier — model releases , new techniques , decisions on what we adopt and what we let the field figure out first Qualities Significant production experience with frontier language models — prompting , provider integration , cost-quality trade-offs at scale Hands-on experience with speech models ( ASR and/or TTS ), including latency , multilingual quality , and provider trade-offs Strong eval discipline — you've built or extended evaluation frameworks , and you know the difference between a benchmark that catches bugs and one that catches drift Comfort with retrieval and grounding ( embeddings, RAG, hybrid retrieval ) and architectural judgment on what belongs in deterministic logic vs language model invocation Strong software engineering background — you ship production code , not just notebooks; TypeScript or Python at depth Track record of shipping production work with AI in your daily loop , and reviewing what it produces Experience pushing voice latency down ( incremental dialogue management , end-of-turn detection ), fine-tuning , dialogue systems , or regulated-domain AI is strongly preferred Fluent in English (B2) and Polish (C1) ; other languages are a plus Why join us Visible impact — your engineering decisions land in real customer experience , not in roadmap slides Real ownership — few layers between you and production, and close work with engineering leadership and founders AI-first workflow with company-paid tooling — Claude Code , model access , IDEs of your choice Flexible contract types ( B2B / Employment Contract ), work arrangements , and working hours Remote, hybrid or in-office — from a location of your choice Direct, low-bureaucracy, results-oriented culture ; no micromanagement Competitive salary , with an honest conversation about trajectory as Lekta scales