Data Scientist
Сербия, Армения, релокация
‍
📍 Location: Serbia, Armenia (We are ready to discuss other countries as well)
🏢 Remote work is possible
💶 Payment terms are open to discussion from 3500 € and up

About the product
At FlameTree, we are building a platform for creating AI agents that help businesses scale customer support, lead follow-up, and sales across multiple communication channels — both inbound and outbound.

Our AI agents work with knowledge bases, communicate in real time, and drive conversions across messaging platforms. The platform supports 150+ languages and integrates with WhatsApp, email, and web applications, offering strong security and high scalability for business growth.

Responsibilities:
— Design and develop the core agent layer responsible for orchestrating interactions with LLMs
— Build and maintain complex conversational logic: state machines, agent workflows, and orchestration pipelines
— Control and shape LLM behavior: prompt design, structured outputs, deterministic flows
— Manage conversational context: memory, history, token limits, and degradation strategies
— Ensure reliability and predictability on top of inherently non-deterministic models
— Implement resilient integrations with LLM providers (timeouts, retries, fallbacks, multi-provider strategies)
— Optimize latency and cost (streaming, batching, caching, token efficiency)
— Debug complex production issues (inconsistent outputs, race conditions, state loss)
— Contribute to system architecture: clear boundaries between agents, backend, and real-time components
— Build observability around LLM pipelines (prompt/response logging, tracing, quality metrics)

Requirements:
— 5+ years of backend development experience with strong Python skills (async, architecture, performance)
— Proven production experience with LLMs (not side projects): understanding of limitations, cost, and behavior
— Experience building agent-based systems or complex orchestration logic (state machines, pipelines)
— Ability to make LLM behavior predictable (structured outputs, schema validation, guardrails)
— Strong debugging skills in non-deterministic systems
— Deep understanding of API integrations (timeouts, retries, idempotency, backpressure)
— Experience optimizing latency and throughput in production systems
— Solid Docker experience and understanding of production environments
— Ability to make architectural decisions independently and take ownership
— Strong engineering mindset: writing maintainable, scalable, production-grade code

Nice to Have:
— Experience with multi-agent systems, tool/function calling
— Experience with local LLMs (Ollama, vLLM, GPU inference)
— Experience with real-time / voice systems
— LLM observability (prompt tracing, evals, quality metrics)
— Cost optimization at scale for LLM usage

What Makes This Role Interesting:
— You will work on the core intelligence layer of the product — not just integrations
— Real production challenges: high load, low latency, reliability requirements
— Direct impact on system architecture and technical decisions
— Fast execution cycle — minimal bureaucracy
— Engineering-driven approach to LLMs (reliability, control, metrics — not just prompt tinkering)
— Strong engineering team focused on building real systems, not prototypes

Who This Role Is NOT For:
— Candidates without real production experience with LLMs
— Engineers relying only on frameworks without understanding underlying mechanics
— Developers without experience in high-load or latency-sensitive systems
— People focused on quick hacks rather than building reliable systems

📩 If you want to join a team where everything is fast, exciting, and truly about AI — drop a message


Понравилась вакансия?
🧡 — да
🌚 — нет


———

💙 Удалёнка
💚 Джунам
💜 За рубежом
🫶 Удалёнка джунам
💚 Качнуть карьеру

мы также есть там, куда всех хотят отправить — https://max.ru/it_vakansii_jobs