Open to opportunities

Patryk Le

@patrykle

Message

Senior AI Engineer building trustworthy RAG systems, production APIs, and data services grounded in customer data.

Poland

Message

What I'm looking for

I’m looking to build and scale production RAG and retrieval services where I can define quality/latency/cost/safety metrics, ship reliable APIs, and strengthen observability and CI/CD—so AI decisions stay grounded, measurable, and trustworthy.

I’m a Senior AI Engineer focused on building retrieval-augmented systems that ground AI outputs in trustworthy customer data. At Help Scout, I spearheaded a 0-to-1 RAG initiative processing 2.3M documents, cutting retrieval latency by 47% and improving answer relevance by 33%, with production-ready standards for ingestion, chunking, and embedding pipelines.

I deliver end-to-end ML features from discovery to launch—defining metrics for quality, latency, cost, and safety, and scaling secure, observable APIs. I built a scalable retrieval service on AWS EKS with Docker and Kubernetes (120k queries/day; 95th percentile latency under 220ms), introduced hybrid search combining FAISS and Elasticsearch (index size -40%, recall +28%), and reduced inference spend by 36% through batching, caching, and cost controls.

Experience

Work history, roles, and key accomplishments

Senior AI Engineer

Help Scout

Sep 2023 - Dec 2025 (2 years 3 months)

Spearheaded a 0-to-1 RAG initiative processing 2.3M documents, integrating OpenAI embeddings with Pinecone to cut retrieval latency 47% and improve answer relevance 33%. Built an AWS EKS-based retrieval service handling 120k queries/day with p95 latency under 220ms and reduced inference spend 36% via batching, caching, and cost limits.

Retrieval Augmented Generation (RAG)Pinecone Faiss fastAPI Embeddings JWT A B Testing

AI/ML Engineer

CGram Software

Jun 2020 - Jul 2023 (3 years 1 month)

Scaled an enterprise logistics search platform using sentence-transformer embeddings and FAISS, achieving 5x higher throughput and reducing average retrieval time to 180ms across 200k documents. Built ETL pipelines for 2TB/month of unstructured data and delivered vector search/classification REST APIs used by 50+ internal clients and 10 external partners.

Sentence Transformers Faiss Airflow Flask fastAPI REST APIs PostgreSQL Amazon S3

Junior ML Engineer

FPT Software

Nov 2018 - Apr 2020 (1 year 5 months)

Developed NLP preprocessing pipelines with spaCy to clean and normalize 400k transport documents, improving downstream model accuracy by 14%. Implemented retrieval prototypes using Elasticsearch and BM25 (72% relevance in pilot tests), and built production ML feature endpoints with Python/Flask and CI testing in Jenkins.

SpaCy Elasticsearch BM25 Python Flask Airflow PostgreSQL Feature Storage Docker Jenkins CI

Data Scientist

TMA Solutions

Sep 2017 - Sep 2018 (1 year)

Built demand forecasting time-series models in Python and scikit-learn, improving forecast accuracy by 18% for logistics planners covering 30 routes. Developed document classification and routing models that reduced manual triage workload 42% and saved ~120 hours/month.

Time Series Python scikit learn MySQL Document Classification Docker Transformer Based Encoders