About Chus
I help companies build AI systems that need to be accurate, fast, and reliable in production. My work focuses on high-impact NLP and LLM problems: improving extraction quality, hardening ML pipelines, evaluating real-world model behavior, and optimizing systems when latency, reliability, or weak evaluation are blocking business value.
I’m most useful when the problem is technically difficult or expensive to get wrong. That includes underperforming NLP systems, fragile AI workflows, unclear evaluation, or inference setups that are too slow or too costly. In one consulting engagement, I rebuilt a PII-masking pipeline, improving global F1 by around 20% while making it roughly 30x faster than the previous setup.
My background combines applied machine learning, research, and systems engineering, so I can work end to end: problem framing, data, experimentation, model and pipeline design, testing, CI, deployment, and optimization. I work well with teams that need strong technical ownership, fast ramp-up, and rigorous execution on complex AI projects.
Spanish
Native or bilingual
English
Native or bilingual
Catalan
Native or bilingual
Remote only
Primarily works remotely
Experience
- AltaMKAI ConsultantLEGALJuly 2025 - October 2025 (3 months)Rebuilt a legal-domain PII masking system from scratch: created a Python framework for preprocessing, multi-entity recognition, resolution, post-processing, and evaluation; designed data strategy, fine-tuned BERT-based models, improved global F1@IoU50 by about 20%, reduced latency about 30×, and deployed on a single RTX 3090 with CI/tests, Docker, and serverless Runpod.
- Georgia Institute of Technology, Systems for AI LabResearch Engineer, Machine Learning SystemsTECHJanuary 2025 - February 2026 (1 year and 1 month)Atlanta, United StatesLed the redesign of Veeksha, an LLM inference benchmarking framework, into a multimodal, session-based framework for DAG-shaped agentic workloads; designed the architecture, configuration system, CI, and a 261-test suite, and guided 4 master’s students.- Contributed to Vajra: built CI and implemented FP8 quantization and DeepSeek V3 / Mixture-of-Experts support across C++ kernels and Python interfaces; validated in tests and inference runs.- Implemented custom telemetry in vLLM and SGLang for advanced prefix caching research; ran Llama 3 70B experiments on 4×H100 (PP2/TP2) and contributed a bug-fix PR to SGLang.
- CaixaBankData ScientistBANKING AND INSURANCENovember 2022 - Today (3 years and 7 months)Barcelona, SpainRedesigned risk models under SAS-only constraints, replacing an external consulting approach and identifying bank guarantee contracts worth about €50 million for cancellation.
Recommendations
Be the first to recommend Chus
Help this freelancer shine by sharing your experience working together.
These freelancer profiles also match your criteria
Agatha Frydrych
Backend Java Software Engineer
4.7
(3)
2
Baptiste Duhen
Fullstack developer
4.6
(4)
5
Amed Hamou
Senior Lead Developer
4
(2)
7
Audrey Champion
Web developer
4.3
(3)
4
Education
- MSc Computer ScienceUniversitat Politècnica de Catalunya2025
- BSc Data Science and EngineeringUniversitat Politècnica de Catalunya2021BSc, Data Science and Engineering. Coursework include: Machine & Deep Learning, Computer Vision, High Performance / Parallel Computing, Video Analysis, Text & Speech Processing, Algorithms or Mathematical Optimization. While performing my studies, I led a team in 2 hackathons: - Winners @ BitsXLaMarato 2020: On the prediction and characterisation of COVID-19 symptoms on the paediatric age. - Runner-ups @ United Nations ICC Global Hackathon: 140 students from 54 global universities participated in the competition. Predicting and segmenting forced population displacements around the world. I was also a student mentor for 2 years and actively participated in the academic comitte.
Skill set
Categories
- Other