Senior Machine Learning Engineer
Company: Textkernel by Bullhorn, Machine intelligence for people and jobs
Location: Amsterdam, Netherlands
Period: Feb 2022 - Present
Domain: HR Tech, High-volume Parsing, NLP & LLM Integration, Microservices.
- Architecture & Modernization
- Standardization: Led the migration of 10+ services to Python 3.11 and Pydantic v2, significantly reducing technical debt.
- Infrastructure: Managed the migration of Docker images for 20+ repositories to AWS ECR, updating Helm charts to align with company-wide standards.
- Monolith to microservices: Drove the extraction of core logic from legacy Perl monoliths into standalone, modern Python services.
- Modular parsing: Designed and implemented a modular pipeline architecture by separating extraction and derivation logic (major refactoring effort), enabling the integration of new features and drastically reducing maintenance overhead, while preserving support for legacy features.
- Performance & Efficiency
- Optimization: Delivered 20–50% performance improvements across key services and reduced the production footprint of high-traffic services by 50%.
- Latency: Reduced complex document processing time from >5s to <160ms by resolving critical tokenizer bottlenecks.
- Resilience: Eliminated recurring outages caused by corrupt input data by implementing robust input validation patterns.
- AI & LLM Strategy
- Knowledge distillation: Used LLMs as teacher models to generate synthetic training data, improving the performance of existing efficient, low-latency production models.
- Data quality: Integrated error-analysis steps into ML training pipelines to ensure consistent output quality.
- Innovation: Co-developed a new LLM-based parsing service as a high-quality alternative engine for in-house CV parsing models.
- LLM Ops: Built an end-to-end evaluation framework to measure LLM cost, latency, and quality to validate production readiness.
- Operational Excellence
- API evolution: Delivered a new version of the Candidate Input API, enabling full and partial ATS data updates.
- Cross-team delivery: Led the end-to-end delivery of a new feature across seven products, coordinating multiple teams to ensure synchronized production deployment.
- Observability: Standardized Grafana dashboards for consistent 4xx/5xx error tracking, speeding up root-cause analysis.
Skills: Python (FastAPI, Pydantic v2), Perl, Docker, Kubernetes (Helm), AWS (ECR), LLMs (Distillation, Parsing, Evaluation), Grafana
Machine Learning Engineer
Company: Textkernel by Bullhorn, Machine intelligence for people and jobs
Location: Amsterdam, Netherlands
Period: Jan 2019 - Jan 2022
Domain: HR Tech, NLP, Skills Intelligence, ML Platform Engineering.
- Service Ownership: Owned the full lifecycle of the “Skills Extraction” microservice, scaling it to handle milions of documents per day across 25+ languages with <50ms (p90) latency. Successfully delivered it as both a standalone product and an integrated component of CV/vacancy parsing.
- Taxonomy & ML Integration: Engineered a hybrid extraction engine combining a strict skills taxonomy with ML-based contextual validation; automated a “feedback loop” (ELK, Jira) that identified “unknown skills” in production traffic and reported them to the Knowledge team, driving continuous taxonomy improvement.
- Complex Parsing Solutions: Engineered a custom parser for PDF LinkedIn profiles and collaborated on the multi-column CV rendering upgrade by designing validation UIs and implementing extraction heuristics.
- CI/CD Standardization: Unified engineering workflows by creating generic GitLab CI/CD templates; drove department-wide adoption where every engineer migrated their service configurations to the new standard, ensuring consistent performance tracking and release automation across the fleet.
- Resilience Engineering: Designed and implemented a standardized internal service client library with built-in retries, timeout policies, and user-friendly error messages. Impact: Adopted by the team for 10+ microservices, significantly stabilizing upstream communication.
- Observability & DevEx: Built internal debug endpoint and Kibana dashboards to trace extraction logic, drastically reducing the “Time to Resolution” for support tickets regarding skill extraction errors.
- DevOps & Support: Ensured platform stability by leading incident response and resolving systematic defects in core parsing products.
- Technical Mentorship: Upskilled the team on Python profiling, CI/CD, and GitOps workflows to foster technical excellence.
Skills: Microservices, NLP, Elasticsearch, CI/CD (GitLab), API Design, Python, Docker, Kubernetes.
Machine Learning Engineer
Company: Qwant, European search engine
Location: Epinal, France
Period: Nov 2017 - Aug 2018
Domain: Search, NLP, Query Understanding
- Query Correction: Worked on automatic query correction for the Qwant European search engine.
- System Design: Designed and implemented a baseline correction system for isolated spelling errors in user search queries, using a two-stage NLP solution:
- generated candidate corrections using edit-distance–based spell checking,
- re-ranked them using a language model.
Source code: ccquery
Skills: NLP, Query Correction, Language Models, Edit Distance, Python (spaCy, fasttext, hunspell, symspell, PyNLPl), SRILM, Docker, GitLab
Machine Learning Engineer
Company: Xilopix, French search engine
Location: Epinal, France
Period: Oct 2016 - Nov 2017
Domain: Machine Learning, NLP, Information Retrieval, Search Engine
- End-to-end ownership: Owned the Machine Learning workflow from raw data acquisition (Elasticsearch) to production deployment.
- Modeling: Built neural network classifiers for webpage topic detection (TF-IDF, LSA) and image color classification:
- trained an LSA model on 40M documents to transform raw text into 300-dimensional semantic vectors.
- achieved 96% F1 score on the held-out dataset.
- Integration: Solved a complex integration constraint by training models in Python and re-implementing inference logic in Ruby, ensuring prediction parity.
- Deployment: Deployed models into production indexing, exposing predictions and probabilities to enable threshold tuning and downstream filtering.
- Engineering rigor: Strengthened software practices around modularity, versioning, testing, performance optimization, and reproducibility.
Source code: xi-ml-topicdiscovery, xi-dip
Skills: Machine Learning, NLP, Text Classification, Neural Networks, TF-IDF, LSA/LSI, Elasticsearch, Python (gensim, scikit-learn), Ruby, Docker, CI/CD
PhD student
Company: Inria (French national research institute), Université de Lorraine
Location: Nancy, France
Period: Dec 2012 - Feb 2016
Domain: Speech Recognition, Language Modeling, Assistive Technologies
- Project RAPSODIE: Conducted doctoral research on building communication aids for the deaf and hearing-impaired. Focused on accurate Speech-to-Text systems for embedded devices with limited memory.
- Hybrid Language Model: Proposed and implemented a novel model combining words and syllables to handle out-of-vocabulary words while preserving accuracy on frequent words.
- Vocabulary expansion: Investigated word similarity based on contextual distributions to support dynamic vocabulary growth in language models.
- Intent detection: Developed a real-time system to detect questions vs. statements using lexical and prosodic features (pitch, n-gram likelihoods), training classifiers like logistic regression, decision trees and shallow neural networks.
Skills: Speech Recognition, Language Models, Feature Engineering, Experimental ML, Distributed Computing, Git, Perl, Java, LaTeX
Support for international employees
Company: Inria, French national research institute
Location: Nancy, France
Period: Feb 2014 – Aug 2015
Facilitated the integration of international researchers by organizing monthly group activities and assisting with administrative procedures.
Junior Research Engineer
Company: Inria, French national research institute
Location: Nancy, France
Period: Oct 2011 - Dec 2012
Domain: Speech Recognition, Applied Machine Learning
- Speech Verification: Developed ML models to detect mismatches between speech and text in non-native speech for language learning systems.
- Feature Engineering: Engineered feature sets comparing constrained (forced) vs. unconstrained (phonetic) text-to-speech alignments, leveraging domain-specific knowledge (phonetic classes, phoneme durations, n-gram likelihoods).
- Analysis: Trained logistic regression models and conducted systematic experiments to analyze the effects of pronunciation variations and training data quality.
Skills: Speech Recognition, Feature Engineering, Logistic Regression, Data Analysis, Model Evaluation
Intern - Research & Software Development
University: Université de Lorraine
Location: Nancy, France
Period: Feb 2011 - Juin 2011
Domain: Speech Recognition, Remote sound, Home Automation
- Optimization: Tested multiple acoustic/language models and decoding configurations to optimize speech recognition in noisy, remote-microphone setups.
- Automation: Developed scripts (Java, Perl, Shell) to automate performance testing across different environments.
Skills: Automatic Speech Recognition (ASR), Remote sound, Java, Perl, Linux
Intern - Research & Software Development
University: Universitatea ‘Stefan cel Mare’
Location: Suceava, Romania
Period: Feb 2010 - Juin 2010
Domain: Human–Computer Interaction (HCI), Gesture-Based Gaming Interfaces
- Prototyping: Engineered a custom setup using a Wii remote and IR LED glasses to capture head position.
- Algorithm Design: Implemented algorithms to filter noise and map natural head motions to video game inputs in real-time.
Skills: C#, Computer Vision, Algorithm Design, Rapid Prototyping, Human-Computer Interaction (HCI)
