Niharika Aggarwal

Built and owned the retrieval pipeline for a RAG-based Product Intelligence tool (Haystack, MongoDB), cutting lookup time from ~10 min to under 30 sec. Developed a full-stack GenAI app (React, FastAPI) with streaming responses, reducing time-to-first-token to under 1.5s. Built a scalable slide retrieval system (Qdrant, vLLM, OCR), cutting ingestion time from 45 to 15 min per 100-slide deck. Dockerized and deployed multiple GenAI apps for reproducible team access.

Built LLM-based PDF-to-JSON extraction tools, converting unstructured safety/regulatory documents into structured datasets and halving manual annotation effort. Fine-tuned open-source LLMs for disaster management, road construction, and traffic regulation use cases. Designed a LlamaIndex-based RAG pipeline in Node-RED, raising answer relevance scores from 3.1/5 to 4.3/5. Applied NLP techniques (chunking, embedding, deduplication) across multiple domains.

Contributed to iQBuddy, a model-recommendation system for Model-Based Systems Engineering (MBSE), leading UI/UX design and prototyping. Curated a labelled dataset of 200+ entries and applied prompt engineering with H2OGPT and ChatGPT-3 to train and evaluate the recommendation logic, measurably improving output relevance.

+ Web Development (Front end) + UI UX + Client Management

Researched classical ML techniques for predictive analytics, contributing to a Coronavirus Prediction System using regression and time-series methods on health data. Collaborated in a 2-member team, owning end-to-end tasks from data preprocessing to model evaluation, and presented findings to the supervising scientist.