Cientista de Dados (IA Generativa)

This position is listed on behalf of a partner company, who manages all applications and next steps. Our partner is looking for a Cientista de Dados (IA Generativa) based in Brazil. This role focuses on designing and building production-ready generative AI solutions for complex business processes. You will architect cognitive engines that transform API inputs into structured, reliable outputs. The position combines LLMs, prompt engineering, multimodal AI, RAG, OCR, computer vision, and automated evaluation. You will work with real-world historical cases to improve model performance, reliability, and decision quality. A major part of the role involves building robust evaluation frameworks and monitoring accuracy, cost, latency, and inference quality. You will collaborate closely with business analysts to validate results and continuously improve AI-driven solutions. The environment is dynamic, technology-focused, agile, and suited to someone who enjoys experimentation, problem-solving, and continuous innovation

The role combines hands-on data science and generative AI engineering with solution architecture, experimentation, evaluation, and production monitoring. You will be responsible for building reliable AI capabilities while ensuring that outputs are measurable, explainable, and aligned with business requirements. Design the architecture of cognitive engines, from API input payloads through to structured JSON outputs. Develop specialized prompts using step-by-step reasoning approaches, anti-hallucination guardrails, and JSON Schema. Implement multimodal analysis using Gemini, including OCR and computer vision for different types of documents. Build structured extraction pipelines for document fields and detect inconsistencies across analyzed content. Design and implement RAG solutions using semantic chunking, metadata by coverage/clause/limit, embeddings, hybrid search, and reranking. Use historical cases, including approved, rejected, and exception cases, to calibrate models through few-shot techniques and edge-case scenarios. Build automated evaluation suites and benchmarks to measure model accuracy, precision, and recall. Implement regression testing whenever prompts or underlying models are changed. Monitor LLM inference quality, cost, and latency using appropriate observability practices. Maintain prompt and experiment versioning. Collaborate with business analysts to validate and homologate AI-generated results. Continuously identify opportunities to improve model performance, reliability, and business outcomes. Requirements The ideal candidate combines strong Python development skills with hands-on experience deploying and evaluating LLM-based solutions in production. You should be comfortable working across prompt engineering, RAG, multimodal AI, evaluation frameworks, APIs, and cloud-based data environments. Advanced Python knowledge and experience developing REST APIs using FastAPI or similar frameworks. Practical production experience with LLMs, including prompt engineering, structured JSON Schema outputs, function calling, and anti-hallucination techniques. Proven experience building RAG solutions, including embeddings, vector databases, chunking, reranking, and retrieval evaluation. Experience with vector databases such as pgvector, Vertex AI Vector Search, Pinecone, Qdrant, or similar technologies. Experience with multimodal models, preferably Gemini, including OCR and computer vision applied to documents and images. Experience designing LLM evaluation methodologies using golden datasets, accuracy/precision/recall metrics, and regression testing. Working knowledge of Databricks, including notebooks, MLflow, and Delta Lake. Knowledge of AWS and ability to collaborate effectively with data engineering teams. Strong Git knowledge and good software development practices, including automated testing. Experience with Google Cloud and Vertex AI, particularly the Gemini API, is a plus. Familiarity with orchestration frameworks such as LangChain, LlamaIndex, or LangGraph is desirable. Experience with evaluation tools such as Ragas, DeepEval, LangSmith, or similar platforms is a plus. Experience in insurance, claims, regulatory processes, or legal and contractual documents is desirable. Experience using MLflow for prompt and experiment versioning is a plus. Knowledge of LGPD and PII anonymization or masking practices is desirable. Experience with LLM guardrails and red teaming is a plus. Databricks, Google Cloud, or AI certifications are desirable. Strong analytical, problem-solving, collaboration, and continuous-learning skills. Benefits PJ contractor arrangement. Anywhere Office model, offering geographic flexibility depending on project requirements. Flexible work environment, with work-location expectations aligned with the specific project and recruiting team. Partner discounts on services such as car rental, language schools, undergraduate programs, postgraduate courses, and other offerings. Talent referral program with rewards for successful referrals. Birthday day off. TotalPass/Wellhub. Telemedicine platform supporting physical and mental healthcare. Health and quality-of-life initiatives focused on employee wellbeing. Access to office spaces designed for relaxation and social interaction, including games, recreational areas, equipped kitchens, and coffee. Dynamic technology environment with exposure to emerging tools and approaches. Opportunity to work on generative AI, multimodal models, RAG, evaluation, and production AI solutions. Collaborative and agile environment focused on continuous learning, experimentation, and innovation.

laptop-ats-crawler v3