
Alan Sammarone
Verified Expert in Engineering
AI and ML Developer
Amsterdam, Netherlands
Toptal member since March 28, 2017
Alan is a lead ML and AI engineer with 17 years of software experience, taking ML systems from prototype to production: multi-modal computer vision on a live seed-sorting machine, real-time model serving on Kafka, and LLM agents, RAG, and generative image and video models. He built and leads the ML team at Enza Zaden after six years working remotely for US companies, and is hands-on from Terraform and Kubernetes to model evaluation and fine-tuning.
Portfolio
Experience
- Python - 12 years
- Linux - 10 years
- PostgreSQL - 10 years
- Machine Learning - 6 years
- PyTorch - 5 years
- Kubernetes - 4 years
- Apache Kafka - 3 years
- Terraform - 3 years
Work Experience
AI Engineering Consultant
Freelance Clients
- Built a RAG system over a corpus of PDFs (under NDA): embedded extracted figures and text with image and text models in pgvector, retrieving related figures and passages as LLM context; measured recall@k on a hand-labeled set.
- Built a LangChain pipeline that turns an uploaded CSV into weak-supervision labeling functions in three LLM stages, each validated against the data, producing training labels without manual annotation.
- Processed geospatial time-series data with xarray for a six-month climate-science engagement, on a dataset in the hundreds of TB handled in slices.
Lead AI Engineer, Tech Lead and Architect
Enza Zaden
- Built and led a 4-person ML engineering and data science team, hiring 3 of them; defined the 3-5 year ML strategy with business stakeholders.
- Built the production serving, evaluation, and live-monitoring infrastructure for a multi-modal seed-sorting model (X-ray, visible light, chlorophyll imaging), recovering €4M per year in otherwise-rejected seed inventory.
- Gated every model release on recall, precision, and PR-AUC against a growing golden dataset of QA germination samples, with drift checks, catching regressions before deployment.
- Held the sorter's inference API to P99.9 latency under 30 ms on an always-on machine that halts when responses slow down, tracing fixes from application code down to on-prem network cabling.
- Built FastAPI-based internal platforms used by AI and biotech R&D teams, on Azure infrastructure managed in Terraform (tens of thousands of resources).
- Built agentic engineering tooling: Claude Code skills and FastMCP servers giving agents read-only access to production logs and databases.
Principal AI + Full Stack Engineer
Nav
- Integrated Tillful's transaction categorization engine, a key asset in Nav's acquisition of Tillful, into Nav's infrastructure after joining through the acquisition.
- Led a 3-person team and owned planning and architecture for moving a large part of the company's data infrastructure from nightly batch to real-time, event-driven Kafka streaming.
- Migrated dozens of batch jobs to streaming hands-on, so ML-driven features reached users in real time instead of next day, reducing customer churn.
Senior ML Full-stack Engineer
Tillful
- Took a financial transaction categorization engine (NLP, weak supervision) from proof-of-concept to production; it became a key asset in Nav's 2023 acquisition of Tillful.
- Built the production ML pipeline for an early-warning default prediction model at one of Europe's largest banks, processing close to 10 TB per run with Spark, Kubeflow Pipelines, and Kubernetes.
- Built Python services (APIs, batch and stream microservices) serving ML models to Fortune 500 companies.
- Acted as principal engineer during the first growth phase, owning architecture (Flask, RabbitMQ, Postgres) as the user base grew from 0 to 20,000 in a year.
- Hired and mentored mid-level engineers; improved automated code quality checks, deployment, and code review processes.
Senior Developer
Simbiose Ventures
- Created a machine-learning pipeline as well as a REST API to access it, which was able to categorize websites according to their contents.
- Optimized various parts of the company's system, generating a 5 to 10-fold performance improvement and decreased costs.
- Migrated the company's storage architecture to a hybrid of Amazon Glacier and Amazon S3, decreasing storage costs by 30%.
- Idealized and oversaw the creation of a system for extracting product prices for any given URL representing a product listing.
Software Developer
Positivo Informática
- Coded highly optimized, browser-based mathematical and physical simulations aimed at helping teaching high school children visualize concepts.
- Wrote an automation tool used to update and deploy thousands of applications efficiently.
- Optimized many legacy JavaScript projects to make them run on low-end tablets.
Junior Developer
Aymará Editora
- Coded and supported a social network aimed at children.
- Wrote JavaScript animations and games for tablets.
- Created a framework to sync information from different databases.
Junior Developer
Totalize Internet Studio
- Created websites for small and medium-sized businesses using a proprietary framework.
- Customized administrative tools and worked closely with engineers focused on securing the platform.
- Communicated with clients to gather requirements and triage bug reports.
Project History
Scalable and Weakly Supervised Bank Transaction Classification
https://arxiv.org/abs/2305.18430I was involved in the ideation and research phases and led the productionization effort. My work included designing and implementing the real-time inference architecture, integrating the system with Kafka and KSQL for event batching, and deploying scalable models via Kubernetes and Kubeflow Pipelines.
The system uses heuristics and unsupervised embeddings to generate weak labels, which are then used to train GRU-based discriminative models. Our pipeline outperformed Plaid API in several tasks, achieving over 90% accuracy across nine financial categories. The architecture was optimized for rapidly onboarding new classification tasks with minimal manual effort.
Voice-controlled Customer Service Agent
Vulkan: AI Image and Video Generation Platform
On the model side, I fine-tune the image and video models and optimize GPU inference with quantization, CPU offload, model caching for cold starts, and spot instance packing.
I run the full production stack: a FastAPI and Postgres back end, NATS job dispatch to spot GPU workers provisioned by a custom control plane, Cloudflare edge delivery with signed URLs backed by R2, Stripe payments, and OpenObserve observability. I also wrote the threat model and security review that gated the public launch.
Claude runs in production with budgets, rate limits, provider fallbacks, and per-call cost tracking, and I wrote most of the codebase through agentic Claude Code workflows.
Education
Master's Degree in Theoretical Physics
University of Amsterdam - Amsterdam
Master's Degree in Computational Science
University of Amsterdam - Amsterdam
Bachelor's Degree in Physics
Universität Leipzig - Leipzig, Germany
Skills
Libraries/APIs
NumPy, SciPy, SQLAlchemy, Pandas, Flask API, React, REST APIs, Asyncio, PyTorch, Stripe, Node.js, Claude API, JAX, ArcGIS, XGBoost
Tools
Pytest, Claude Code, Claude, GitHub, Amazon Simple Queue Service (SQS), Amazon Elastic Container Service (ECS), Terraform, Apache Airflow, Celery, RabbitMQ, Helm, Grafana, GitLab CI/CD, AWS CloudFormation, Amazon CloudWatch, Figma to HTML, Git, Microsoft Excel, ComfyUI
Languages
Python, SQL, Python 3, GraphQL, JavaScript, TypeScript, PHP, C, Rust
Frameworks
Flask, Django, Streamlit, FastMCP, Apache Spark, Next.js
Paradigms
Real-time Systems, DevOps, Test-driven Development (TDD), REST, Web Application Architecture, Event-driven Design (EDD), Load Testing, Search Engine Optimization (SEO), Model Context Protocol (MCP)
Platforms
Amazon Web Services (AWS), Docker, Kubernetes, Apache Kafka, Azure, Raspberry Pi, Twilio, Google Cloud Platform (GCP), Cloud Run, Vercel, Amazon, Linux, Kubeflow, Webflow
Storage
PostgreSQL, Database Modeling, MySQL, Redis, NoSQL, Elasticsearch, Aerospike, Google Cloud
Industry Expertise
Trading Systems
Other
Machine Learning, APIs, API Integration, Large Language Models (LLMs), Technical Leadership, Software Architecture, Web Development, Artificial Intelligence (AI), Machine Learning Operations (MLOps), CI/CD Pipelines, AI Engineering, AI Model Training, Back-end, Agentic AI, Infrastructure as Code (IaC), Architecture, OAuth, SaaS, Startups, API Gateways, Back-end Development, Document Parsing, Data Extraction, AI-enabled Applications, Real-time Data, Real-time Data Pipelines, Full-stack Development, Cloud Engineering, System Architecture, Scalability, API Design, Agentic Coding, Full-stack, LLM Agents, Fintech, Analytics, Data Modeling, Product Engineering, Web App Development, Computer Vision, Data Scraping, Algorithms, Web Scraping, Data Science, FastAPI, Cython, OpenAI, Data Visualization, Data Analysis, AI Agents, Azure Blob Storage, Distributed Systems, Cloud Architecture, Retrieval-augmented Generation (RAG), LangChain, Prompt Engineering, Statistical Modeling, Voice AI, DSP, Cloud Monitoring, AI Automation, LLM Integration, MLflow, AI-assisted Development, Information Retrieval, PDF, Finance, Prometheus, Application Security, Generative Engine Optimization (GEO), Front-end Design, User Experience (UX), User Interface (UI), Web UI, Web UX, Consulting, Payment Gateways, Mathematical Analysis, Apache Cassandra, Neural Networks, Physics, Spectroscopy, Chrome Extensions, Trading, Leadership, Ideation, Delivery, Biology, Genomics, Containers, Model Monitoring, Model Deployment, Natural Language Processing (NLP), Pgvector, Xarray, Geospatial Data, Generative AI & LLM Fine-Tuning, Cloudflare, NATS, Design
How to Work with Toptal
Toptal matches you directly with global industry experts from our network in hours—not weeks or months.
Share your needs
Choose your talent
Start your risk-free talent trial
Top talent is in high demand.
Start hiring