MHP – A Porsche Company

- Developing multimodal Retrieval Augmented Generation applications on company data with Large Language Models
- Building an LLM-based Quality Evaluator and ML-based Duplicate Detection Method for Software Bug Tickets

Data Scientist

Bosch Power Tools

- Implementation and evaluation of probabilistic deep learning models for demand forecasting on AWS SageMaker

Master Student Data Science

- Development of a tool for automated data analysis and visualization in SAP Analytics Cloud
- Implementation of time series forecasting and outlier detection models

Working Student Data Science

Mercedes-Benz AG

- Speeding up the assembly process by using machine learning to predict throughput times and errors on Azure Databricks
- Automation of graph visualization for the optimization of assembly analysis

Fraunhofer IPA

- Implementation and evaluation of a generative deep learning model for image data augmentation for cutting edge patterns
- Implementation of scientific papers in Python

Bachelor Student Deep Learning

MAHLE

- Analysis of engine data and application of machine learning for prediction in piston temperature measurement

Intern Data Analysis & IT Project Management

I make learning Data Science effortless for you | Data Science @ Bosch

Hi 👋🏻
I'm Banias, Data Scientist at MHP.

Applying new technologies and methods in Data Science is my passion.

Through my experience and continual learning, I've gained valuable insights and tips that I love sharing with Data Scientists, Machine Learning Engineers and builders like you.

If you're looking for practical advice, helpful resources, and articles on the latest trends or best practices in the field, you're in the right place.

Let's connect and stay updated on the exciting world of Data Science together. Feel free to reach out, read my articles, and let's grow our knowledge and network. :)


Note: My views are my own and don‘t represent my employer.

Banias Baabe

Vitraya Technologies

Question Answer using LLAMA 3.1 8B and OpenAI API:

•Developed an advanced NLP solution using OpenAI's 4o-mini API to automate claim validation by extracting
and evaluating policy criteria from insurance documents, achieving 95% accuracy and help client in saving
money.
•Work extensilvely on desgining the right prompt.
•Have done preprocessing of data to reduce the false positive.
•Currently we are getting around 95% of accuracy on this.

Clinical attribute extraction from document:

•Developed LLM based applications for the insurance domain, replacing traditional rule-based approaches with
advanced deep learning methods.
•Fine-tuned LLAMA 2 8B for precise attribute and value extraction from documents, enhancing data accuracy
and processing efficiency.
•Have reduce hallucination caused by the model by using chain of prompt and fine tunning.
•Worked on fixing the output structure. I used post processing, Prompt engineering and fine tunning for this.

Procedure Classification from Preauth Document:

•Built a comprehensive 8 class classification solution for disease identification, effectively managing data that
falls outside these classes using Logistic regression with 89% Precision and 91% recall.
•Have using confidence score threshold to handle classes which is not the part of training data.
•used Lime interpretor for figuirng out the keywords that was responsible for giving incorrect response. and wrote a preprocessingcode to remove those keywords to further imporve the accuracy of the model.

Fixing bugs and doing Enhancement:

•Implementing multiple bug fixes and improvements to significantly boost product stability and accuracy

Software Engineer - NLP

American Express

- Worked on multiple feature to devlop the MLOPS platform for the Data science team.
- Created a Python based Library that helped in reducing time by 70% for creating configuration file to deploy the Model.

Analyst - Data Science

Subex

- Using Dask library for paralleling the machine learning task and saving time. On multiple algorithms, I have achieved a 50% reduction in time.

- I build a Name entity recognition model. Precision is 81% and recall is 78%. I have created the pipeline using Kubeflow.

- Created a proof of concept for Dask Optuna. So that we can parallelize the auto hyperparameter optimization for various models. We have achieved a 23% reduction in time.

- I helped the team in annotation correction for improving the model accuracy.

- Building api using fastapi for deep learning models.

- Working on dynamic resource allocation to machine learning model based on the data size and the model.

Cart4baby

- Creating end to end plan for the product journey.

- Running Ad campaign through adwords and Instagram.

- Did the SEO and almost all of our pages got ranked in top 3 on Google.

- Managed a team of 10 people.

Market Analyst

BYJU'S

Business Development Trainee

"Over the past year, I've worked on various LLM-based projects, fine-tuning models like LLAMA 2 (8B) to achieve optimized results for specific use cases. I also have hands-on experience utilizing the GPT API directly." In my experience working with LLMs, I've encountered three significant challenges: false positives, hallucinations, and dynamic output structures. Through persistent effort, I've developed solutions to overcome these issues, ultimately making the product ready for production.



--- Data scientist @subex ---

As a data scientist at Subex, I played a key role in developing a NO code AI platform. I achieved significant milestones, including building a Name Entity Recognition model with an impressive 82.1% F1 score and creating a pipeline using Kubeflow. 

Additionally, I utilized the Dask library to parallelize machine learning tasks, reducing training time by 30% on large datasets. I also successfully implemented Dask-Optuna for automating hyperparameter optimization in parallel, resulting in a 23% time reduction. 

Furthermore, I contributed to annotation correction, improving model accuracy, and developed APIs using FastAPI for deep learning models. Lastly, I focused on dynamic resource allocation to optimize machine learning models based on data size and model requirements.

--- M.Tech @ IIIT Allahabad in Data science ---

During my Masters, I conducted a research project focused on extracting patient age from clinical trial data. Remarkably, we achieved a high F1 score of 72%, surpassing the performance of other works in the field. Notably, even the esteemed IBM Research Lab had a 10% lower score than our approach, highlighting the superiority of our methodology.

Furthermore, I contributed to an image caption project where I implemented a hybrid approach to enhance model accuracy. Through the utilization of a knn model, I was able to achieve accuracy levels comparable to human performance. Impressively, our model attained a BLEU-4 score of 0.19, further validating the effectiveness of our approach in generating accurate image captions

taplio

The Battle

Banias Baabe

Inzamam Safi

Followers

Posts

Engagements

Likes

Ready to grow your LinkedIn presence?