Biniyam Ajaw Odol

Biniyam Ajaw Addis Ababa , Ethiopia - |- Data Scientist and ML Engineer linkedin.com/in/biniyamodol/ | github.com/biniyam69 Professional Profile Not just passionate as in a dreamer but an actual passionate ML engineer who is willing to put in the hard work that is required by the industry. Always moving along with the industry by learning, coding, and solving community problems with a strong background in Deep Learning, math, and programming. Going along with the state-of-the-art technologies with RAGs, NLP,, Computer Vision and Generative AI. Education 2023 Adama Science and Technology University Relevant courses: • Web Development • Applied Mathematics • Kernel Development • Natural Language Processing • Algorithms • Operating Systems • Realtime Embedded systems Skills Python, Rust, Scala, SQL FastAPI, React, Nodejs, Langchain.js, Expressjs. Projects OpenAI API, LangChain, LangSmith, RAGAS, TruLens, PineCone, Cohere, Rag Evaluation, SDXL, Lora, QLoRA, HuggingFace API, Model Inference, RLHF, SFT • Project Management • Mobile App Development Visualization ML / Deep Learning Tableau, Redash, Transformers, Regression, Classification, CNN, RNN, LSTMs, RLHF, Reinforcement Learning, Attention Models, Fine Tuning, Pretraining. Gen AI Programming/Statistics GPA: 3.45 Bsc in Computer Science Others Git, AWS, Azure, Pytorch, Tensorflow You can check out my projects on my github here. Some of the projects are Redash Chat Plugin to allow visualization of data with Natural Language, Automated Storyboard Generation for Advertisements with Dall-e and Stable Diffusion, AI Contract Lawyer RAG to help people understand and draft contract papers, Geo Ethereum Dapp that refunds by location of drivers. Fine tuning quantized open source LLMs with LoRA. Fine tuned Mistral and LLama 2 7B on Amharic Dataset to help generate Ads in Amharic. Experience ML Engineer / AI Trainer March 2022 - June 2023 Purpose Black Addis Ababa, Ethiopia Model Fine tuning with LoRA and Peft for ideal ideal code completions for C, Python, and the natural language of Amharic Model Performance Enhancement: Continuously evaluate and improve the outputs generated by AI models, actively addressing biases, errors, and shortcomings. Collaborate closely with researchers and engineers to refine the models and achieve higher quality results. Data Curation and Pipeline Design: Curate diverse and relevant datasets to train AI models, ensuring representative and comprehensive coverage. Design and implement efficient pipelines for data processing, augmentation, and model training to maximize training effectiveness. Bias Identification and Mitigation: Vigilantly identify and address biases within AI-generated responses, actively working to minimize any unfair or unintended consequences. Collaborate with diverse teams to promote fairness, inclusivity, and ethical considerations in AI model outputs. Data Scientist April 2018 - August 2021 ML Engineer September 2018 - Jan 2019 Backos Technologies Adama, Ethiopia (Hybrid) Game Analytics and Metrics: Led the development and implementation of a cutting-edge game telemetry system for an upcoming cross-platform game, codenamed "Battle of Adwa: Triumph of Ethiopia." The system is expected to capture comprehensive player data and analyze gameplay patterns to inform the game's design and mechanics after the launch of the game. Predictive Modeling and Player Behavior Analysis: Employed advanced deep learning techniques to create powerful predictive models for player behavior within the game. Conducted extensive research and analysis on player segmentation, churn prediction, and personalized recommendation systems, aiming to enhance player engagement and satisfaction during gameplay. A/B Testing and Experimentation: Devised a robust A/B testing framework to measure the potential impact of different game features and UI enhancements in the forthcoming "Battle of Adwa" game. Analyzed experiment results to identify optimal level designs and character customization options that are expected to boost player satisfaction and overall retention rates during the game's testing phase. Cross-functional Collaboration: Actively collaborated with a team of talented game designers, developers, and product managers to align data-driven insights with the envisioned gameplay experience for the game. St Paul Hospital Hybrid, Addis Ababa Developed an ML model with 95% accuracy for early detection of breast cancer and brain tumors. The model leveraged advanced algorithms and deep learning techniques, trained on a large dataset of medical images. Implemented robust data preprocessing and feature engineering techniques, improving model performance by 20%. By applying state-of-the-art techniques for data cleaning, normalization, and feature extraction, the model's ability to differentiate between normal and abnormal tissue structures was enhanced. Achieved industry-leading performance with a validation accuracy of 97% and sensitivity of 92% for breast cancer and brain tumor detection. Rigorous validation and evaluation protocols were established to assess the model's performance using large-scale datasets. Collaborated closely with medical researchers, enabling them to reduce diagnosis time by 30% and make more accurate treatment decisions. By actively engaging with medical professionals, understanding their requirements, and incorporating their expertise, the developed ML model effectively supported their diagnostic processes.