Skip to main content

Resume

Three ATS-friendly variants, plus the JSON Resume schema. Pick a tab and download the PDF, or copy the canonical URL below for the structured data.

Marcelo Prates, PhD

AI Engineer

Summary

Senior AI/ML engineer with 7+ years across data science, ML engineering, and applied AI, plus PhD research experience. Builds and ships production GenAI and ML systems across React/Next.js frontends, FastAPI and TypeScript backends, retrieval/reranking, agent runtimes, evaluation, and operational reliability.

Skills

GenAI and LLM SystemsOpenAI APIs · Azure OpenAI · LlamaIndex · Agentic RAG · ReAct Tool-Using Agents · DSPy · PydanticAI · MCP Integration · Tool Calling and Orchestration · Structured Outputs · Metadata Filtering · Hybrid Retrieval and Reranking · Memory and Context Management · Vercel AI SDK
AI Stack EcosystemLangChain · LangGraph · OpenAI Agents SDK · CrewAI · Semantic Kernel · Hugging Face Transformers · Pinecone · Qdrant · ChromaDB · FAISS · MongoDB · Convex
Full-Stack Product EngineeringNext.js 15 · React · TypeScript · Tailwind · Clerk · FastAPI · SSE Streaming · REST APIs
Data and Retrieval InfrastructurePostgreSQL · pgvector · Supabase · Redis · Elasticsearch · SQL · Vector Search
ML and EvaluationPython · PyTorch · Scikit-learn · MLflow · Bayesian Optimization · Model Evaluation · Experimentation · Regression QA
Platform and DeliveryDocker · Kubernetes · CI/CD · OpenTelemetry · Prometheus · Grafana · Vercel · DigitalOcean App Platform
System Design.

Experience

Lead Engineer

HopHR, Remote, Brazil

  • Led Moon's production candidate-search platform, unifying retrieval, scoring, reranking, and explainable output so ambiguous hiring requests could become auditable shortlist decisions.
  • Implemented backend APIs, async workflows, and runtime infrastructure on Supabase/PostgreSQL with Redis-backed caching and queues to support reliable, high-throughput hiring operations.
  • Improved release safety through CI/CD hardening, test-driven refactors, and stronger observability across search, streaming, and agent-runtime components.
  • Designed production AI search architecture combining Elasticsearch filters, semantic retrieval, multi-criteria reranking, and LLM-based assessments to improve shortlist quality while keeping results inspectable by recruiters.
  • Built chat-driven hiring workflows with planning, tool orchestration, memory-aware context injection, and AI SDK-compatible SSE streaming across backend and product surfaces.
  • Designed and operated a PydanticAI-centric multi-agent runtime: specialist profiles, delegated subtasks, memory-aware shortlist re-evaluation, and feedback-triggered workflow adaptation.
  • Built persistent LLM memory and session-aware context-injection infrastructure for candidate-search learnings with confidence-scoped retrieval across sessions.
  • Strengthened runtime observability with request correlation, trace propagation, safe logging/redaction, and metrics/tracing patterns suited for debugging long-running agent workflows.
  • Stack: TypeScript, Next.js 15, React, Tailwind, Clerk, Supabase, PostgreSQL, Redis, Elasticsearch, OpenAI APIs, PydanticAI, Vercel AI SDK.
  • Developed model-to-filter translation that turns learned SHAP/feature-importance signals into executable Elasticsearch filter clauses.
  • Partnered with recruiters and hiring managers to convert ambiguous role requirements into testable search/ranking hypotheses.

Lead Data Scientist

Dataside, Remote, Brazil

  • Promoted to Lead Data Scientist; mentored team members and owned client-facing discovery, solution design, and technical delivery across multiple accounts.
  • Standardized evaluation, documentation, deployment, and stakeholder-alignment routines so ambiguous business problems became measurable ML roadmaps, data requirements, and KPIs.
  • Designed and shipped agentic RAG systems with tool-augmented retrieval, stateful memory, and graceful degradation for autonomous enterprise knowledge assistants (Azure OpenAI, LlamaIndex ReAct workflows, PostgreSQL/pgvector).
  • Built research-assistant and ranking workflows with dynamic question generation, document-scoped search, hybrid retrieval modes, rank fusion, and source-aware answers for autonomous answer quality.
  • Hardened the retrieval layer with safe vector-store initialization, advisory-lock-based schema setup, embedding-dimension safeguards, and non-destructive defaults for live indexed data.
  • Designed autonomous fallback paths and regression QA workflows — CLI probes and Playwright-backed browser validation — so assistants remained usable when orchestration dependencies or credentials were unavailable.
  • Delivered full-stack agentic applications with FastAPI backend and React frontend, deployed to Azure Container Apps with Grafana observability and Docker Compose orchestration.
  • Delivered unsupervised and decision-science solutions including clustering/segmentation, anomaly detection with SHAP-based explanations, experimentation/A-B testing, and hybrid NLP pipelines combining TF-IDF, BM25, embeddings, UMAP, HDBSCAN, and vector search.
  • Architected production ML pipelines with PySpark, SQL, Azure, Databricks Jobs, Azure DevOps CI/CD, MLflow model registry, containerized train/predict workflows, and scheduled retraining or model-renewal loops.

AI Engineer

SingularityAI (Part-Time), Remote, Brazil

  • Led end-to-end development of AI voice assistant with real-time transcription, TTS generation, and persona-aware RAG responses.
  • Built YouTube→ChromaDB pipeline with AssemblyAI transcription, speaker diarization, and metadata-rich chunking for knowledge-base extraction.
  • Implemented LangGraph agents with PostgreSQL checkpointing for persistent conversation state across sessions.
  • Integrated ElevenLabs TTS, AssemblyAI STT, and Ultravox for real-time voice conversation with audio event handling (interruptions, pauses).
  • Designed PCA-based scope analysis and concurrent 22-attribute persona extraction for personalized, personality-adjusted responses.
  • Built MinIO-synced ChromaDB vector store with Redis caching for production-ready retrieval infrastructure.

LLM Consultant

Vortigo (Part-Time), Remote, Brazil

  • Built internal assistant chatbots grounded on proprietary code and spreadsheet knowledge sources.
  • Delivered retrieval and prompt pipelines with production guardrails for repeatable workflows.

Senior Data Scientist

Condati, Remote, USA/Brazil

  • Led redesign of bid-optimization and forecasting ML systems for digital marketing workflows.
  • Built Python and SQL pipelines for production forecasting, anomaly detection, and decision support on Snowflake and Aurora-backed marketing data.
  • Partnered with business stakeholders to align model behavior with campaign KPIs and operational constraints.

ML / Computer Vision Consultant

ConstructIN, Remote, Brazil

  • Led production computer-vision delivery for construction monitoring and analytics.
  • Coordinated model development, validation, and delivery workflows with client teams.

Senior AI Researcher (ML for Health)

Samsung Research Brazil, Campinas, Brazil

  • Led wearable-health ML feature development deployed globally in Galaxy Watch products.
  • Delivered memory-efficient inference designs for constrained devices and real-time operation.
  • Coordinated cross-functional research-to-product delivery with HQ stakeholders.

Data Scientist

Poatek IT Consulting, Porto Alegre, Brazil

  • Delivered DS/ML projects across optimization, NLP, computer vision, and risk modeling.

Education

PhD in Computer Science

Federal University of Rio Grande do Sul (UFRGS)

Canonical: marcelo prates.github.io/#resume