Husain Abbas

Dedicated Data Analyst and AI Engineer with a proven ability to leverage data for improved business operations and informed decision-making. Skilled in developing machine learning models and GenAI applications, passionate about utilizing AI to drive innovation and make a meaningful impact. Committed to staying up-to-date with the latest advancements in the field of AI.

1. Classification ML model in Azure - 

• Created a pipeline in Azure that can train, register, and deploy an ML model that can recognize the difference between tacos and burritos.
• Set up Kubeflow in AKS cluster managed by Kubernetes, preprocessed the data, trained the model, registered it to ACR and deployed the model to AML.

2. Customer Churn Prediction with SageMaker Studio XGBoost Algorithm -

• Using Gradient Boosted Trees to predict mobile customer departure.
• Built a model that predicts whether a customer is likely to churn, and then how to optimally set a threshold that accounts for the cost of true positives, false positives, and false negatives.
• Built, trained, tuned and deployed the machine learning model in AWS Sagemaker.

3. Understanding Network Usage Similarities using AI  (Master Thesis)-

• Developed an unsupervised machine learning model, i.e a clustering model to group the different 4g network base stations according to their data usage similarities.
• Prepared the raw data by performing steps like data cleaning, feature engineering, data parameterization and data transformation.
• Performed time series clustering using various algorithms like Time Series KMeans, KShape and Global Alignment Kernel. Also, compared the results for choosing the most optimum algorithm.

4. Gen AI - LLM Project - Q/A Chatbot -

• Built a system that can talk to MySQL Database.
• User asks questions in a natural language and the system generates answers by converting those questions to an SQL query and then executing that query on MySQL database.
• The database consists of inventory, sales aBuilt a hands-on system to fine-tune an open-source LLM on Azure ML, enabling natural-language Q&A over a custom corpus (e.g., product manuals, policy docs).nd discounts data stored in a MySQL database.
• In the UI, store manager will ask questions in a natural language and it will produce the answers. The web app will open in your browser where we can ask questions.
• Technologies used - Google Palm LLM, Hugging Face embeddings, Streamlit, Langchain, Chroma DB.

5. Gen AI – LLM Fine-Tuning on Azure ML – Domain-Specific FAQ Bot

• Built a hands-on system to fine-tune an open-source LLM on Azure ML, enabling natural-language Q&A over a custom corpus (e.g., product manuals, policy docs).
• Users submit questions via a simple web UI, and the back end serves the fine-tuned model as an API endpoint for ultra-fast, domain-aware responses.
• Technologies used – Azure ML, MLflow, PyTorch, Hugging Face Transformers, Git, Docker, Kubernetes.

6. Gen AI – RAG System – Llama 3 Knowledge Assistant

• Developed a Retrieval-Augmented Generation pipeline using Llama 3 and Chroma DB. The assistant ingests PDF and Markdown docs into a vector store, then uses LangChain to retrieve context and generate precise answers to user queries.
• Perfect for building intelligent help desks or research assistants.
• Technologies used – Llama 3, LangChain, Chroma DB, Python, Git, Docker, Streamlit.

7. Gen AI – Multi-Agent AI System – Autonomous SQL Analyst

• Built a collaborative agent framework where distinct AI “experts” handle SQL translation, query optimization, and result analysis.
• Uses AutoGen to orchestrate agents that convert natural-language analytics requests into SQL, execute them on Azure PostgreSQL, and summarize insights back to the user.
• Technologies used – Azure OpenAI GPT-4, LangChain, AutoGen, Azure PostgreSQL, Docker, Kubernetes, Git.

8. Gen AI – Few-Shot Reasoning Benchmark Hub

• Aggregated a suite of complex reasoning tasks (math, symbolic, factual QA) to benchmark LLM performance under few-shot settings.
• Automated evaluation across datasets like MATH, BBH, HumanEval, and MMLU, providing clear visual comparisons of model capabilities.
• Technologies used – Python, PyTorch, CUDA, Tslearn (for time-series tasks), Git.