Candy Aho

– Designed and developed a pedagogical platform powered by a specialized LLM (GPT-40). – Built an adaptive engine for diagnostics, learning plans, intelligent quizzes, and final reporting. – Used LangChain, Streamlit, FastAPI; integrated LLM inference pipeline with evaluation. – Automated creation of personalized learning paths based on student profiles and answers.

– Worked on RLHF and prompt engineering to align multimodal models with human intent. – Contributed to dataset labeling and interpretability efforts for safer, more transparent outputs. – Applied ethical AI practices in data curation and evaluation. – Improved visual recognition accuracy by 25% through data and model refinement. – Collaborated cross-functionally to optimize model performance.

– Fine-tuned Donut model for information extraction from bank checks and receipts. – Applied advanced techniques (LoRA, QLORA, MTLORA, distillation, pruning): -56% inference cost, -71% GPU usage, -74% memory consumption. – Deployed CI/CD pipelines with experiment tracking using MLflow. – Contributed to technical architecture design and documentation.

– Developed LSTM model for stock price prediction. – Built complete data processing pipeline with real-time visualization using Streamlit. – Enhanced analytical capabilities through real-time prediction dashboards.