Senior Data Scientist
- Verfügbarkeit einsehen
- 0 Referenzen
- auf Anfrage
- 81475 München
- auf Anfrage
- pt | en | es
- 15.01.2025
Kurzvorstellung
- adtech
- IoT
- Sensor Data
- Logistics
- Optimization
Qualifikationen
Projekt‐ & Berufserfahrung
7/2024 – 11/2024
Tätigkeitsbeschreibung
AI-Driven Creative Simulation & Optimization (Classic ML, Generative AI, Tree Models, Causality)
• Built a system to integrate multimodal data (tags, chapters, audio, text) and optimize creative arrangements to maximize conversion rates.
• Trained a LightGBM model (Python) to predict conversions using multimodal inputs, including video tags (Grounding DINO, SAM) and audio embeddings (Wav2Vec).
• Designed and implemented a linear programming optimizer using PuLP to simulate optimal configurations under constraints (e.g., budget, runtime).
• Leveraged branch-and-bound and integer programming techniques for optimization, ensuring high accuracy and scalability.
• Deployed and validated the system with live A/B testing, using FastAPI for model serving and AWS EKS for container orchestration.
Generative KI, Marketing-Analytiker, Natural Language Generation, Natural Language Processing, Scikit-learn, Systems Simulation
7/2024 – 11/2024
Tätigkeitsbeschreibung
Generative AI for Multi-Modal Tagging (Text & Image, Generative AI, Fine-tuning, Qwen 2 VL)
• Fine-Tuned Models using Chain-of-Thought reasoning: Created a domain-specific dataset by pairing output from DINO/SAM (visual segments) with corresponding text descriptions.
• Utilized LoRA (Low-Rank Adaptation) to fine-tune model layers efficiently on GPUs with constrained memory, reducing fine-tuning time by ~40%.
• Employed mixed precision training (PyTorch AMP) to accelerate training without sacrificing model accuracy.
• Validated fine-tuned models with BLEU and ROUGE-L scores to ensure domain relevance and coherence in outputs.
• Quantization: Performed Post-Training Quantization (PTQ) and 8-bit dynamic quantization to reduce GPU memory consumption by ~35% with minimal performance loss (<2% accuracy drop).
• Prompt Engineering & Validation: Implemented a prompt chain approach with LangChain, ensuring consistent label generation across varied ad categories (fashion, automotive, tech).
• Deployment: Exposed model endpoints via AWS ECR (containers) on EKS, using HPA (Horizontal Pod Autoscaler) to spin up GPU-enabled pods on demand.
Big Data, Data Science, Data Scientist, Generative KI, Langchain, Natural Language Generation, Natural Language Processing, Natural Language Understanding, Prompt Engineering
7/2024 – 11/2024
Tätigkeitsbeschreibung
Ad Fatigue Predictor (Scikit-Learn, Numpy, Pandas, Matplotlib, Plotly, Survival Analysis)
• Modeled time-to-event data using Python’s Lifelines (Kaplan-Meier & CoxPH), integrated with a Scikit-learn pipeline for hyperparameter tuning.
• Partitioned the performance table in PostgreSQL by time intervals (weekly), speeding up queries ~30%.
• Built a custom Datadog integration to trigger Slack alerts whenever hazard ratios exceeded thresholds.
Data Science, Python, Python-Programmierer, Scikit-learn, Statistiken, Statistischer Analyst
7/2024 – 11/2024
Tätigkeitsbeschreibung
Video Segmentation & Keyframe Detection (Computer Vision, Python, Scikit-Learn)
• Employed PyTorch and OpenCV and scikit-learn to detect scene boundaries (color histograms, fade/cut transitions) and generate keyframes.
• Created FFmpeg-based Docker containers for standardized video encoding/decoding pipelines.
• Scaled batch jobs on AWS Batch, storing intermediate frames on S3 and metadata in PostgreSQL.
Amazon Web Services (AWS), Docker, Postgresql, SQL
6/2024 – 11/2024
Tätigkeitsbeschreibung
Image Segmentation & Tagging (Computer Vision, Python, Pytorch, Gen AI)
• Packaged Grounding DINO and Segment Anything Model (SAM) into separate Docker images; deployed on AWS EKS with KEDA to auto-scale pods based on incoming media volumes.
• Created custom PyTorch inference scripts to handle large-batch GPU processing, using CUDA streams for concurrency.
• Merged CV outputs (bounding boxes, masks) with campaign metrics in PostgreSQL for cross-analysis (e.g., “Which object correlates with higher CTR?”).
Computer Engineering, Computer Vision, Generative KI, Machine Learning, Machine Learning Engineer, Reinforcement Learning
1/2022 – 8/2024
Tätigkeitsbeschreibung
AdWords Analytics (Looker, LookML)
• Developed LookML explores and views for standardized metrics (CTR, ROAS, CPA), enabling consistent cross-team reporting.
• Leveraged Looker Data Actions to push real-time budget updates back into ad platforms via webhooks.
Data Science, Data Scientist, Generative KI, Pytorch, Python-Programmierer
1/2022 – 8/2024
Tätigkeitsbeschreibung
AdWords Infrastructure Setup (Kubeflow, Python, Docker, Kubernetes)
• Built dbt transformations to unify Google Ads data (impressions, costs, conversions) in PostgreSQL.
• Created Airflow DAGs to push and pull data from internal platform and external partners, having to work in data reconciliation and established SLAs.
Data Science, Big Data, Postgresql, Data Scientist, Python-Programmierer, Data Analyst
1/2022 – 8/2024
Tätigkeitsbeschreibung
Probabilistic Attribution with Incrementality Testing (Python, Spark, Causal Inference)
• Developed probabilistic clustering models in PySpark to attribute conversions to campaigns using incomplete fractional data.
• Applied causal inference methods (e.g., Double Robust Learners via EconML) to estimate the incremental lift of campaigns.
• Built incrementality testing frameworks to evaluate the effectiveness of A/B-tested campaigns and non-exposed control groups.
Data Science, Python-Programmierer, Python, Amazon Web Services (AWS), Marketing-Analytiker
7/2020 – 1/2021
TätigkeitsbeschreibungCreated a Mean-Variance Optimization engine (PyPortfolioOpt) for a multi-asset portfolio, reducing overall volatility by ~7.6%.
Eingesetzte QualifikationenPytorch, Python-Programmierer
Ausbildung
Portugal
Über mich
Weitere Kenntnisse
Persönliche Daten
- Portugiesisch (Muttersprache)
- Englisch (Fließend)
- Spanisch (Gut)
- Deutsch (Grundkenntnisse)
- Französisch (Grundkenntnisse)
- Europäische Union
- Schweiz
Kontaktdaten
Nur registrierte PREMIUM-Mitglieder von freelance.de können Kontaktdaten einsehen.
Jetzt Mitglied werden