Senior ML/AI Engineer
⚲ Remote
23 000 - 34 000 PLN (B2B)
Wymagania
- AI
- Python
- Neo4j
Opis stanowiska
Wymagania:
- 5+ years in ML engineering / applied AI / search & retrieval / backend AI systems- Strong Python — production services, not just notebooks- Hands-on with LLMs: retrieval, embeddings, structured outputs, tool/agent orchestration- Experience building evaluation pipelines (golden cases, automated checks, human review loops)- Understanding of retrieval failure modes: stale context, hallucination amplification, ranking drift- Knowledge graphs / graph-based memory (Neo4j or similar)- APIs, async jobs, observability, schema validation, production debugging- Strong bias toward measurable quality improvements over demo-only progress- AI-native dev tools: Claude Code, Codex, or similar
Nice to Have
- Fine-tuning, distillation, or local model deployment- Experience with sovereignty-driven model strategy- Agentic systems or long-term memory for AI products- Pragmatic reasoning across vector / lexical / graph / hybrid retrieval approaches
Codzienne zadania:
- Build scalable memory and retrieval systems for high-context AI workflows
- Hybrid retrieval across knowledge graphs, vector, lexical, and structured approaches
- Evaluation workflows comparing retrieval and memory strategies on real corpora
- Agent memory interfaces — enabling AI workflows to query, inspect, and reuse context
- Quality metrics: recall, provenance, latency, cost
- Routing logic — deciding when to use lightweight retrieval, deeper reasoning, or agentic workflows
- Fine-tuning, distillation, and model adaptation pipelines
- Observability for AI workflows: retrieval failures, model regressions, cost budgets
- 5+ years in ML engineering / applied AI / search & retrieval / backend AI systems- Strong Python — production services, not just notebooks- Hands-on with LLMs: retrieval, embeddings, structured outputs, tool/agent orchestration- Experience building evaluation pipelines (golden cases, automated checks, human review loops)- Understanding of retrieval failure modes: stale context, hallucination amplification, ranking drift- Knowledge graphs / graph-based memory (Neo4j or similar)- APIs, async jobs, observability, schema validation, production debugging- Strong bias toward measurable quality improvements over demo-only progress- AI-native dev tools: Claude Code, Codex, or similar
Nice to Have
- Fine-tuning, distillation, or local model deployment- Experience with sovereignty-driven model strategy- Agentic systems or long-term memory for AI products- Pragmatic reasoning across vector / lexical / graph / hybrid retrieval approaches
Codzienne zadania:
- Build scalable memory and retrieval systems for high-context AI workflows
- Hybrid retrieval across knowledge graphs, vector, lexical, and structured approaches
- Evaluation workflows comparing retrieval and memory strategies on real corpora
- Agent memory interfaces — enabling AI workflows to query, inspect, and reuse context
- Quality metrics: recall, provenance, latency, cost
- Routing logic — deciding when to use lightweight retrieval, deeper reasoning, or agentic workflows
- Fine-tuning, distillation, and model adaptation pipelines
- Observability for AI workflows: retrieval failures, model regressions, cost budgets
🔍 Dekoder Ogłoszenia
🟡
Strong Python — production services, not just notebooks
Oczekuje się, że kandydat będzie tworzył stabilne i skalowalne usługi w Pythonie, a nie tylko eksperymentował w środowiskach deweloperskich.
🟡
Strong bias toward measurable quality improvements over demo-only progress
Priorytetem są realne, mierzalne wyniki i poprawa jakości, a nie tylko prezentacje działających prototypów.
🟡
Fine-tuning, distillation, or local model deployment
Może oznaczać, że firma chce optymalizować istniejące modele lub wdrażać je w sposób bardziej efektywny, ale niekoniecznie musi to być główny obszar odpowiedzialności.
🟡
Pragmatic reasoning across vector / lexical / graph / hybrid retrieval approaches
Oczekuje się umiejętności elastycznego podejścia do różnych metod wyszukiwania danych i wybierania najodpowiedniejszej w danej sytuacji.
🟢
Agentic systems or long-term memory for AI products
Może wskazywać na zaawansowane projekty związane z autonomią systemów AI i ich zdolnością do zapamiętywania i wykorzystywania informacji w dłuższym okresie.