ADAMCZYK#07
★ 2026 Series · Limited Edition

SzymonAdamczyk

Machine Learning Engineer

An all-round player who turns research into production. I design, train and deploy ML systems end to end, from the first notebook to monitored, autoscaled services that handle millions of predictions a day.

0Years Pro
0+Models Shipped
0Overall Rating
Sign the Player Watch Highlights
Click card to flip · Move to tilt
01

Player ProfileScouting Report

Szymon is a Machine Learning Engineer who plays both sides of the ball: rigorous enough to read the papers, pragmatic enough to ship what works. Over eight seasons he has built recommendation engines, forecasting systems, computer vision pipelines and, more recently, retrieval-augmented LLM products used by thousands of people every day.

His game is built on fundamentals: clean data contracts, reproducible training, honest evaluation and observability that catches drift before customers do. He is most valuable in the gap between data science and platform engineering, turning promising prototypes into reliable, cost-efficient services.

Off the field he mentors junior engineers, writes internal playbooks on model evaluation, and stays close to the research frontier in efficient inference and evaluation of LLM systems.

— Verdict: franchise-level starter

A+

Production ML

Training pipelines, model serving, CI/CD for models, canary rollouts.

A

LLMs & Retrieval

RAG, embeddings, fine-tuning (LoRA), evaluation harnesses, guardrails.

A

Applied Modeling

Ranking, forecasting, tabular boosting, deep learning for vision and NLP.

A−

Data Engineering

Spark, streaming features, feature stores, data quality checks.

A

Team Play

Clear writing, cross-team delivery, mentoring, product thinking.

02

ExperienceSeason Stats

2023— NOWLIVE

Senior Machine Learning Engineer

Nordlight Analytics · Remote, EU
  • Leads the ML platform squad: built a retrieval-augmented assistant over 2M+ internal documents, now used by 6,000 employees.
  • Designed an LLM evaluation harness (offline golden sets plus online feedback) that gates every prompt or model change.
  • Cut GPU serving cost by 41% with quantization, batching and autoscaling on Kubernetes.
14Models in prod
−41%Serving cost
6KDaily users
4Engineers mentored
2020— 2023

Machine Learning Engineer

Vistula Retail Tech · Łódź, PL
  • Shipped a real-time product ranking service (two-tower retrieval plus gradient-boosted re-ranker) serving 30M requests per day.
  • Built a demand forecasting system for 120K SKUs that reduced stock-outs and cut forecast error by 27%.
  • Introduced MLflow, feature store and drift monitoring to standardize the model lifecycle.
+8.4%Conversion lift
38msp99 latency
−27%Forecast error
30MRequests / day
2018— 2020

Data Scientist (Rookie Season)

Kestrel Industrial AI · Łódź, PL
  • Developed a CNN-based visual defect detector for a manufacturing line, deployed to edge devices.
  • Built predictive maintenance models on sensor time series, reducing unplanned downtime.
  • Automated data labeling workflows with active learning, halving annotation time.
97.2%Defect recall
6MBEdge model size
−50%Labeling time
ROYRookie of the Year
2016— 2018

Draft Class: M.Sc. Computer Science

Specialization: Machine Learning · Poland
  • Thesis on sample-efficient image classification with semi-supervised learning.
  • Research assistant in a university ML lab; top-10 finish in two Kaggle competitions.
1stRound pick
2×Kaggle top 10
M.Sc.Degree
MLMajor
03

ProjectsCareer Highlights

HL #01LLM · RAG

Knowledge Desk Assistant

Retrieval-augmented assistant with hybrid search (BM25 plus embeddings), a cross-encoder re-ranker, citation grounding and automated hallucination checks.

PyTorchpgvectorvLLMFastAPI
0.91Recall@10
−63%Tickets
1.2sMedian
HL #02RECSYS

Real-Time Ranking Engine

Two-tower candidate retrieval with ANN search, followed by a LightGBM re-ranker on streaming features. Rolled out with interleaving and A/B tests.

TensorFlowFAISSKafkaRedis
+8.4%Conv.
38msp99
30MReq/day
HL #03FORECASTING

Demand Forecast Grid

Hierarchical forecasting for 120K SKUs combining gradient boosting with a temporal fusion transformer and reconciliation across stores and regions.

SparkLightGBMAirflowTFT
−27%WAPE
120KSKUs
−18%Stock-outs
HL #04VISION · EDGE

Defect Spotter

Quantized, distilled CNN that detects surface defects at line speed on low-power edge hardware, with active learning to keep the model improving.

PyTorchONNXTensorRTOpenCV
97.2%Recall
6MBModel
45FPS
HL #05MLOPS

Model Factory Platform

Self-service training and deployment platform: templated pipelines, model registry, shadow deployments and automatic drift alerts for every team.

KubeflowMLflowTerraformPrometheus
5d→4hDeploy
9Teams
99.97Uptime
HL #06LLM EVAL

Eval Referee

An evaluation harness for LLM features with golden datasets, pairwise model-graded scoring calibrated against human labels, and CI regression gates.

PythonDuckDBGitHub ActionsLoRA
0.86Human agr.
2.4KTest cases
0Silent regr.
04

SkillsThe Stat Table

CAREER SKILL TOTALS · CLICK HEADERS TO SORT
Skill Category Seasons Projects Rating
PythonModeling842
97
PyTorchModeling724
95
Model Serving & InferencePlatform619
93
LLMs, RAG & Fine-tuningModeling39
92
Kubernetes & DockerPlatform621
90
Recommender SystemsModeling46
91
MLflow · Kubeflow · AirflowPlatform517
89
SQL & Data ModelingData838
90
Spark & Streaming (Kafka)Data512
86
Computer VisionModeling47
86
Time-Series ForecastingModeling58
88
AWS · GCP · TerraformPlatform620
85
Experimentation & A/B TestingData6190
89
CAREER3 CATEGORIES8 SEASONS40+ MODELSOVR 94
05

ContactSigning Desk

FREE AGENT · OPEN TO OFFERS

Draft him to
your roster

Building ML products, scaling a platform, or need a hand getting an LLM feature production-ready? Get in touch directly. Replies are usually within one business day.

Send Email Call Now
Name
Szymon
Surname
Adamczyk
Address
Uniwersytecka 4
95801 Zgierz
Poland
Phone
+1 (810) 537-1375
Email
gildermayallreciqu37@gmail.com