Consultant
I embed with product and engineering teams to design, build and ship AI/ML and data systems end to end. Recent clients include The RealReal, Paramount, Offerpad and FlatWorld.
I turn AI and data into reliable products.
I'm André Claudino, a senior engineer who works across the stack: generative AI and LLM agents, machine learning, backend and distributed systems, data and cloud infrastructure. For over a decade I've taken ideas from prototype to production at scale, and kept them fast, reliable and cost-efficient. I work as a consultant, author, speaker, professor and creator of advanced software.
Consultant · Author · Speaker · Professor · Creator · AI/ML · MLOps · Backend
13+
years shipping software
8
years ML & MLOps in production
10M+
peak requests / minute
−87%
infra & inference cost cut
Consultant, author, speaker, professor and creator
My work takes several shapes, from hands-on delivery to teaching and building in the open. These are the roles I bring to a team.
Author
I write long-form technical articles on MLOps, generative AI, Rust for ML, data and FinOps, and I publish open-source software and libraries.
Speaker
I speak at technology events including Python Brasil and Python Rio, bringing production lessons to engineering and business audiences.
Professor
I taught at the State University of Rio de Janeiro (UERJ) and designed and coordinated a graduate Machine Learning & MLOps course (Unyleya), mentoring engineers on high-performance systems and production ML.
Creator of advanced software
I build and maintain demanding software: typed-lm, sciencekit, serde-onnx, k8s-maestro and IndieSync, an AI audiobook production platform.
Breadth across the stack, depth where it matters
I work end to end: from product idea and architecture to production operations. My background spans software engineering, machine learning, generative AI, data and infrastructure, which lets me own a problem from the first prototype to scale.
In plain terms: I help companies use AI and data to make better
decisions, faster and cheaper, and I build the software and infrastructure that makes
those systems dependable.
Generative AI & LLM agents
Autonomous agents, RAG, tool calling and MCP, fine-tuning and domain-specific small models. Built an analytics agent that lets executives query data lakes in natural language (report time from ~3 weeks to immediate, ~77% cost reduction).
Machine learning in production
Recommender systems serving peaks of ~10M requests/minute, fraud and risk, churn, LTV and VIP detection, pricing and optimization, and computer vision.
Backend & distributed systems
High-throughput, low-latency services and APIs in Rust, Go and Python, event-driven architectures and messaging. Rewrote an NLP serving engine from ~10 minutes to ~10ms per request.
MLOps & platform engineering
Kubernetes ML platforms, CI/CT/CD for models, registries, monitoring, drift detection and retraining. Redesigned a platform to cut infrastructure costs by ~87%.
Data engineering & big data
Lakehouse architectures, distributed computing and streaming with Spark, Trino and Kafka, feature stores and reliable ingestion pipelines.
Cloud & FinOps
AWS, GCP and Azure with Terraform. I treat cost as a first-class requirement, with a track record of up to ~87% reduction in infrastructure and inference spend.
flowchart LR
src["Data sources<br/>events · APIs · DBs"]:::neutral
lake[("Data lakehouse<br/>Trino · Spark · S3")]:::data
feat["Feature store<br/>offline + online"]:::data
train["Training<br/>Argo · Kubeflow · MLflow"]:::ai
reg["Model registry<br/>versioned artifacts"]:::ai
serve["Serving<br/>K8s · Knative · Triton"]:::system
obs["Monitoring & drift<br/>OpenTelemetry"]:::infra
client["Clients<br/>APIs · apps"]:::actor
src --> lake --> feat --> train --> reg --> serve --> client
serve --> obs
obs -. retrain .-> train
classDef actor fill:#fef3c7,stroke:#f59e0b,color:#78350f,stroke-width:1.5px
classDef system fill:#dbeafe,stroke:#2563eb,color:#0c4a6e,stroke-width:1.5px
classDef data fill:#d1fae5,stroke:#059669,color:#064e3b,stroke-width:1.5px
classDef ai fill:#ede9fe,stroke:#7c3aed,color:#3b0764,stroke-width:1.5px
classDef infra fill:#e4e4e7,stroke:#71717a,color:#18181b,stroke-width:1.5px
classDef neutral fill:#f4f4f5,stroke:#a1a1aa,color:#18181b,stroke-width:1.5px
Models are software.
They need tests, observability, rollbacks and cost control, not just good metrics.
What I work with
A decade of production engineering across AI, backend, data and infrastructure, with a bias toward performance and low latency. Rust, Python and Kubernetes are my daily tools.
Backend & Distributed Systems
High-throughput, low-latency services and APIs, event-driven architectures, messaging and streaming. Rust, Go and Python, with an emphasis on reliability and performance.
Machine Learning & Generative AI
LLMs (RAG, fine-tuning, prompt policy), autonomous agents, tool calling, MCP, NLP, computer vision, recommender systems, small language models and ONNX optimization.
MLOps & Platform
Enterprise MLOps infrastructure: Kubernetes, Knative, Argo Workflows, Kubeflow, CI/CT/CD for ML, shift-left validation, model versioning, monitoring and retraining loops.
Data & Cloud
Lakehouse architectures, distributed computing and streaming, plus cloud-native scalability and FinOps across AWS, GCP and Azure.
Languages & Optimization
Rust, Go, Python, Scala and SQL, with a focus on ONNX optimization, low-latency API design and high-concurrency execution.
Where I've built things
Employment and consulting engagements, most recent first.
Senior Machine Learning Engineer
The RealReal
07/2025 – present
Consulting · Remote
- Owns pricing ML end to end at a luxury consignment marketplace: LightGBM price models (BPM/LPM/LDM).
- Serving through Triton and a gRPC price-estimation API, with a Rust port of the serving path for performance.
- MLOps on GCP Vertex AI, contrastive image embeddings for taxon understanding, and observability across the pricing stack.
AI/ML Consultant
Offerpad
04/2026 – 07/2026
Consulting · Remote
- Architected an automated real-estate appraisal system assisted by computer vision, integrating classical ML with visual LLMs (Gemini, Claude).
- Codified expert underwriting knowledge into reproducible ML pipelines, minimizing individual bias.
- Deployed and optimized models with ONNX, moving humans to a reviewer role and shrinking the end-to-end workflow to minutes.
AI/ML Consultant
Paramount
12/2025 – 03/2026
Consulting · Rio de Janeiro
- Designed the core MLOps architecture for distributed initiatives: recommendation engines, automated image tagging, preference learning and LLM-assisted search.
- Integrated AI-assisted shift-left validation into CI/CD, building end-to-end testing agents that codify team knowledge.
Staff AI/MLOps Engineer
Stellar Gaming (EstrelaBet)
01/2024 – 04/2026
- Autonomous GenAI analytics agent: an LLM agent letting C-levels query data lakes in natural language; report time from ~3 weeks to immediate, ~77% cost reduction.
- High-volume recommendation & betting: a recommender serving peaks of ~10M requests/minute during live events, plus an NLP "Tipster" platform for natural-language sports bets.
- Cost optimization & SLMs: trained domain-specific small language models and redesigned the Kubernetes ML platform, cutting infrastructure costs by ~87%.
Manager, Data Engineering & ML
Kognita Labs
04/2022 – 10/2024
- Architected a high-performance ML/Data platform on Trino, Argo and Kubernetes, achieving a 56% cloud infrastructure cost reduction via dynamic scaling and Spot instances.
- Rewrote a legacy NLP inference service's multi-threaded serving engine in Rust and optimized models via ONNX for CPU: latency from 10 minutes to ~10ms, operating cost −87%.
Head of Data & ML (Consultant)
FlatWorld.co
05/2022 – 03/2023
- Built a Lakehouse (Trino/GCP) from scratch, consolidating fragmented spreadsheet data for real-time analytics.
- Built explainable ML models generating transparent audit records for algorithmic recruitment, cutting inference match time from days to real-time.
Head of Data & ML
Talentify.io
03/2021 – 03/2022
- Created a platform for fine-tuning text foundation models on specific role domains and built conversion-based candidate matching models.
- Re-architected monolithic role classification into event-driven, multi-model solutions on Kubernetes.
Machine Learning & MLOps Manager
B2W Digital
08/2018 – 03/2021
- Pioneered MLOps adoption within Latin America's largest e-commerce company, merging software engineering and data-science workflows.
- Architected a high-performance freight optimization model and predictive systems that anticipate API failures ~10 minutes in advance.
Software Engineering Specialist
Stone Pagamentos
10/2017 – 02/2018
- Spearheaded an architectural transformation for back-office big data processing, resolving cascading failures across on-premise nodes.
- Containerized and orchestrated the environment with Marathon, enforcing strict CPU/memory boundaries for financial transaction processing.
Software Engineer Specialist
B2W Digital
04/2018 – 08/2018
Graduate Coordinator, ML & MLOps
Unyleya
06/2018 – 03/2019
- Built, coordinated and managed the team of a Machine Learning and MLOps course focused on high-performance systems.
Software Development Analyst
WebRadar
07/2016 – 07/2017
- Developed and improved a complex suite of applications and APIs for processing and visualizing big data for the telecommunications market.
Senior Software Engineer
Tecgraf / PUC-Rio
07/2013 – 03/2016
- Planning, scheduling, demand management, customer contact and testing for large systems (e.g. Petrobras ERP).
Software Engineering Technical Leader
K2 Sistemas
07/2013 – 03/2016
- Led software development in spatial data, real-time demand and embedded software.
flowchart LR
offline["Offline eval<br/>golden sets · metrics"]:::ai
gate["CI/CT gates<br/>quality thresholds"]:::system
canary["Shadow / canary<br/>traffic split"]:::system
online["Online monitoring<br/>latency · cost · quality"]:::infra
drift["Drift detection<br/>data & concept"]:::infra
retrain["Retrain trigger<br/>automated pipeline"]:::ai
offline --> gate --> canary --> online --> drift --> retrain
retrain -. feedback .-> offline
classDef system fill:#dbeafe,stroke:#2563eb,color:#0c4a6e,stroke-width:1.5px
classDef ai fill:#ede9fe,stroke:#7c3aed,color:#3b0764,stroke-width:1.5px
classDef infra fill:#e4e4e7,stroke:#71717a,color:#18181b,stroke-width:1.5px
Things I build in the open
Rust-first projects across ML, LLM systems, Kubernetes tooling and developer agents. Most live under the neurono-ml org.
🧭
typed-lm
Single-forward-pass semantic routing in Rust: turns dense LLMs into a typed decision API: booleans, choices and scores instead of generated text. Built on Candle, with LoRA/QLoRA training and FP8/FP4 quantization.
🧪
sciencekit
A classical machine learning library, similar to scikit-learn, written and optimized in Rust. Zero-copy APIs, pluggable execution (CPU/SIMD/BLAS/GPU) and native out-of-core processing, with Python bindings via PyO3.
📦
serde-onnx
Serde-first ONNX intermediate representation in Rust with a generated protobuf codec and typed ai.onnx.ml operator bindings. Pure Rust, no C/C++ dependency.
☸️
k8s-maestro
Type-safe Kubernetes workflow orchestration in Rust. Define multi-step workflows with a fluent API: DAG dependencies, parallelism, retries, checkpointing, RBAC and network policies.
🐇
rabbitmq_streamer
Rust library wrapping RabbitMQ publish/consume behind streams: high-level APIs, manual acks, automatic reconnection and exponential backoff, tested against real brokers.
🥒
cucumber-openspec
Deterministic converter from OpenSpec spec.md behavior specifications into Cucumber/Gherkin .feature files in 80 languages, with zero AI. Keeps specs and tests in sync via CI.
🧰
kind-testcontainers-skill
AI-agent skill that scaffolds ephemeral Kubernetes integration tests with Testcontainers and Kind: isolated cluster per test, kubeconfig port-mapping and CI templates across 7 languages.
🦀
rust-dev-agents & leptos-coder
OpenCode agents and skills for Rust development: ML/DataFusion, Actix-web APIs, Tonic gRPC, Leptos web apps, Kubernetes deployment and Docker optimization.
🎙️
IndieSync
AI-assisted platform that turns a manuscript into a full-cast audiobook/podcast with consistent character voices, SFX and music. Event-driven pipeline (Rust + Python, Qwen3-TTS), broadcast-ready mastering. Private beta.
flowchart LR
user["Executive<br/>natural language"]:::actor
agent["Analytics agent<br/>LLM orchestrator<br/>tool calling · MCP"]:::ai
sql["SQL / Trino<br/>query tool"]:::tool
rag["RAG<br/>vector store"]:::tool
code["Python<br/>analysis tool"]:::tool
guard["Guardrails<br/>access · cost · safety"]:::infra
lake[("Data lake")]:::data
answer["Answer + charts<br/>auditable"]:::actor
user --> agent
agent --> sql --> lake
agent --> rag
agent --> code
sql --> guard
rag --> guard
code --> guard
guard --> answer
classDef actor fill:#fef3c7,stroke:#f59e0b,color:#78350f,stroke-width:1.5px
classDef ai fill:#ede9fe,stroke:#7c3aed,color:#3b0764,stroke-width:1.5px
classDef tool fill:#fee2e2,stroke:#ef4444,color:#7f1d1d,stroke-width:1.5px
classDef data fill:#d1fae5,stroke:#059669,color:#064e3b,stroke-width:1.5px
classDef infra fill:#e4e4e7,stroke:#71717a,color:#18181b,stroke-width:1.5px
Academic background
A physics foundation with a focus on artificial intelligence and symbolic computation, plus ongoing specialization in data and AI engineering.
Ph.D. in Computational Physics
AI & Symbolic Computation, State University of Rio de Janeiro (UERJ), 2013–2016.
M.Sc. in Computational Physics
Symbolic NLP Frameworks, UERJ, 2010–2012.
B.Sc. in Physics
Distributed Big Data Research, UERJ, 2007–2010.
Postgraduate Specialization in Data Science
PUC-Rio, 2025–2026.
MBA in AI Engineering
Full Cycle, 2025–2026.
MBA in Software and Systems Architecture
Full Cycle, 2025–2026.
Let's build reliable AI systems
I'm open to senior AI/ML, MLOps and forward-deployed engineering roles. If you need someone who can take a model from a notebook to production and keep it healthy, let's talk.