I’m an aspiring AI engineer with a strong interest in large language models (LLMs), retrieval-augmented generation (RAG), and trustworthy AI systems. I enjoy building applied ML projects that sit at the intersection of theory and real-world impact, especially around transparency, reliability, and human-centered AI.
I’m particularly interested in:
- Large language models and prompt engineering
- Retrieval-augmented generation (RAG) systems
- Reducing hallucinations and improving model reliability
- Model evaluation, interpretability, and responsible AI
- Applied ML systems and MLOps workflows
This GitHub serves as a space to document my learning, experiments, and end-to-end projects as I work toward becoming an AI engineer.
A real-time American Sign Language (ASL) translation system that uses computer vision and hand tracking to recognize signed letters and convert them into text.
Key ideas & features:
- Real-time hand detection and landmark tracking using MediaPipe
- Multiple implementations (web-based, rule-based, and ML-powered) to explore different design tradeoffs
- Rule-based gesture recognition for interpretable, geometry-driven classification
- Neural network–based models (TensorFlow / PyTorch) for improved accuracy and scalability
- End-to-end ML pipeline including data collection, preprocessing, training, and inference
- Low-latency, live translation with extensible support for custom datasets and new signs
This project explores the intersection of computer vision, human–computer interaction, and machine learning, with a focus on accessibility, real-time systems, and comparing symbolic versus learned approaches to perception.
An end-to-end RAG chatbot designed to answer questions grounded in a specific document corpus rather than relying solely on model memory.
Key ideas & features:
- Document ingestion, chunking, and vector storage
- Semantic retrieval to surface relevant context
- LLM-based answer generation constrained by retrieved sources
- Explicit handling of “I don’t know” cases to reduce hallucinations
- Emphasis on evaluation, citation, and reliability
This project reflects my interest in trustworthy LLM applications, particularly in how retrieval, prompting, and evaluation can be combined to improve factual accuracy.
A recommendation system that goes beyond keyword matching by leveraging semantic embeddings to understand user intent and content similarity.
Key ideas & features:
- Uses embedding-based similarity to recommend books by meaning, not just keywords
- Supports flexible, natural-language user queries
- Focuses on explainability (why a book was recommended)
- Designed as a modular ML pipeline (data → embeddings → retrieval → ranking)
This project explores how representation learning can improve traditional recommender systems while remaining interpretable and user-focused.
- Hallucination detection and confidence scoring for LLM outputs
- Model interpretability dashboards for ML/LLM systems
- MLOps practices for deploying and monitoring AI applications
- AI applications in high-stakes domains (e.g., healthcare, education)
If you’re interested in AI systems, LLM applications, or responsible ML, feel free to explore my repositories or reach out. I’m always excited to learn, collaborate, and get feedback.
Thanks for stopping by!