Harshita Mandalika

I've built end-to-end applied AI systems, from hybrid RAG pipelines and agentic LLM workflows to multimodal prediction models.

My RAG evaluation dashboard combines ChromaDB dense search, BM25 retrieval, and Reciprocal Rank Fusion across 574 PubMed abstracts, reaching 0.90 RAGAS faithfulness. I also built an e-commerce analytics agent that translates natural-language requests into safe SQL and chart recommendations across 541K+ transactions.

For a product feedback platform, I turned 10K+ app reviews into prioritized insights using NLP classification, clustering, Gemini API summaries, FastAPI, React, and Docker—reducing manual review effort by 80%. I've also fine-tuned Mistral 7B with QLoRA and PEFT for consumer health summarization, improving ROUGE-L from 17.2% to 60.8%.

I'm completing a Master's in Computer Science at Texas A&M and bring research experience in supervised machine learning, including a first-authored dengue fever prediction publication. I enjoy building measurable, deployable AI products with Python, PyTorch, LangChain, LangGraph, and cloud tooling.