QUEST LOG: ML & NLP DEPLOYMENTS

Building a lightweight denoising model as a pre-processing layer to make RAG systems robust against typos and OCR errors. Designing a comprehensive benchmark to evaluate real-world resilience. Master's Thesis — Ongoing.

Hybrid ensemble of ML models (Logistic Regression, SVM, Random Forest, XGBoost) + LLMs (Llama2/3) for suicide risk detection from social media text. Achieved 95% accuracy.

Semantic book recommendation engine using Sentence Transformer embeddings, LangChain, and ChromaDB. Gradio dashboard with Zero-Shot Classification and Sentiment Analysis.