Building Soda to monitor from data to decision

I love talking about: machine learning, data quality, decision making, post deployment data science, performance estimation, and bunch of other stuff. 

Always open to have a chat 🙂

Hakim Elakhrass

KAYA AI

●      Translated business requirements to model and system metrics
●      Collaborated closely with cross-functional teams to ensure a balance between performance, cost, and deadlines.
●	Led multi-modal dermatology data acquisition and analysis, ensuring a stratified and reproducible dataset split. 
●      Built a robust data preprocessing pipeline, reducing processing time from 1 hour to 10 minutes.
●      Prototyped a DL model for diagnosis, applying architecture insights from research papers, improving AUPRC by 0.3 and reducing training time by 7 hours.
●      Led the design and implementation of scalable pipelines on AWS for model training and inference with GPU optimisation and data parallelisation.
●	Conducted fine-tuning and optimisation of foundational models with 4-bit QLoRA for treatment plans generation, significantly reducing model loading and inferencing time.
●      Steered the conversion of notebooks into a production-ready codebase, ensuring modularity and maintainability.

Machine Learning Engineer

* Single-Cell Genotoxic Response Modeling:
● Researched and applied insights from academic publications on DNA damage to understand its mechanisms and connections with genetic instability, genotoxicity testing, cancer therapies, and the evolution of degeneration in postmitotic neurons. 
● Used Mask R-CNN to segment AP sites in PBMCs, and adopted it for real-world deployment, handling custom datasets and MLOps workflows. 
● Implemented OOP principles and managed project dependencies with Poetry. 
● Handled large-scale data for genotoxicity experiments, ensuring reproducibility and scalability using Feast, DVC, and Backblaze. 
● Ensured transparency by versioning models and experiments through MLflow and Databricks CE, allowing reproducibility and traceability of model behavior and decision-making. 
● Deployed Docker images and utilized GitLab CI/CD with pre-commit hooks for continuous integration and delivery. 
● Ensured a secure storage of secrets by using Gitlab CI/CD variables. 

* Computational Biology and Molecular Modeling: Coursework and self-directed study to build foundational knowledge in single-cell RNA-seq, protein structure and interaction networks, and AI applications in these fields. 
●     Sanbomics’ Single-cell RNA-seq analysis. 
●     EMBL’s Protein interactions and their importance. 
●     EMBL’s Approaches for using protein protein interaction networks for biological discovery. 
●     UC Berkeley’s Deep Learning for Protein Engineering Primer (AlphaFold 2, ProteinMPNN, RFDiffusion).

Independent Researcher in ML for Biology

HubForward

●	Integral member of the frontend team, quickly adapting to React and contributing code within two weeks before transitioning to API development and software design.
●	Top contributor with over 1000 commits, demonstrating a strong commitment to project success.
●	Co-led the containerization of the application using Docker, which eventually improved performance and ease of installation by 40%.
●	Managed the deployment to DigitalOcean and led the migration to DigitalOcean Spaces which sped up file I/O by 2s.
●	Co-managed the backlog to ensure well-defined user stories, leading to a 15% increase in sprint goal achievement.
●	Participated in peer code reviews with backend and frontend teams, resulting in a 20% increase in team cohesion.
●	Implemented TDD practices using PHPUnit, leading to a significant reduction in production bugs by 35% due to early identification. 
●	Mentored an intern who became a full-time employee within 1 month.
●	Co-authored and maintained comprehensive technical and non-technical documentation, facilitating clear communication and knowledge sharing within the team and with stakeholders.
	Tackled challenges with changing application requirements by predicting future changes based on past patterns and designing a scalable architecture, reducing major updates by 20%.
●    Another challenge I faced was managing time as I was working full-time and studying at the same time. This allowed me to really improve my time management skills in order to not affect my studies, work, or my colleagues.

Software Developer

I.R Chic

*   Started building the web app from scratch till a shippable version where I learned the WAMP stack, Scrum, and Security best practices. 

*   Took the lead on Scrum adoption, refined the Product Backlog, estimated user stories, broke them down into tasks, and planned sprints. This eventually cut down development time by 50%.

*   Followed Git and GitHub best practices by creating a new branch for each user story I was going to be working on, implemented it, ran manual Unit and Integration Tests to insure its proper functioning, and then merged into the main branch. In case there was a bug, I would study its severity and effect on UX to decide when to address it. I also had the most Git commits and merge requests in the team with over 300 commits.

*   Followed OWASP's security and testing guides like PHP prepared statements, strong password policies, and anti-DDoS protection using Google Recaptcha.

*   I was also in charge of the server configuration where I configured sending emails, file uploads, and ensuring only the proper errors and warnings were enabled.

Backend Developer

I'm a software developer turned ML Researcher/Engineer translating AI into impact for human health.

Most recently, I led the development of an AI assistant for dermatology diagnosis and treatment planning - covering everything from data acquisition and preprocessing to model prototyping, optimisation, and cloud deployment.

I’ve also built and deployed a deep learning pipeline for DNA damage analysis in single-cell electrophoresis assays, enabling predictive insight into compound toxicity.

Currently, I’m focused on deepening my expertise in SOTA AI/ML architectures and how they can be used to advance healthcare and biotech. I also contribute to a weekly newsletter where we break down the latest tools in life science AI.

I’m driven by curiosity and the desire to change the world positively. Always happy to connect with like-minded people!

taplio

The Battle

Hakim Elakhrass

Abdelkrim Laloui

Followers

Posts

Engagements

Likes

Ready to grow your LinkedIn presence?