Nitish — Data Analyst/Scientist

About

Turning data into
decisions.

Nitish Rohilla

Business Analyst

I'm a Data Analyst/Scientist based in Gurugram, India. currently working at PolicyBazaar. I specialize in end-to-end ML/AI solutions — from Data Insights, Professional Dashboards and NLP pipelines to cloud deployments on AWS and GCP.

I hold a B.Tech in Computer Engineering (Data Science) from J.C. Bose University, YMCA Faridabad. I'm driven by the challenge of turning messy, real-world data into clear, actionable insights and Implementing the Power of machine learning algorithms on it.

Outside of work, I contribute to open-source game mods, build personal projects, and always have something new in the queue to learn.

Get in touch

Core Skills

Python

SQL

ML / Deep Learning

LLM Fine-Tuning / RAG

JavaScript / C++

1yr+

Professional ML experience

Projects shipped

Cloud platforms (AWS, GCP, n8n)

Experience

Where I've worked.

April 2026 - Present

PolicyBazaar.com

Business Analyst

Preparation of Large Datasets from Raw Data and end-to-end design and deployment of production-grade Data Insights, Interactive Dashboards driving measurable business impact.

AWS Athena/QuickSight: Leveraged AWS Athena for Preparation and Testing of the Datasets and Implementing those dataset in AWS Quicksight to prepare High Quality Dashboards and Extract Valuable insights from the Data

March 2025 — March 2026

SkillUpTech

Associate Data Scientist

Spearheaded end-to-end design and deployment of production-grade AI/ML solutions, driving measurable business impact across NLP, LLM fine-tuning, and intelligent automation.

n8n AI Agent Automation: Architected multi-agent workflows with deep expertise in SQL Queries, JSON schema design, agent chaining, and tool/memory integration. Built a fully autonomous Password Reset Agent with secure OTP generation and token-based authentication.

AWS SageMaker: Leveraged SageMaker for scalable spaCy model training and Lambda for serverless inference endpoints — enabling cost-efficient, auto-scaling deployments.

GCP Vertex AI — Gemini Fine-Tuning & RAG: Fine-tuned a Gemini model on Google Cloud Vertex AI for domain-specific tasks. Designed a RAG agent integrating embedding models and vector search — significantly reducing hallucination rates.

Projects

Things I've built.

+5% BERTScore F1

Qwen 2.5 LLM Fine-Tuning with QLoRA

Fine-tuned a Qwen 2.5 0.5B instruction model on local hardware using QLoRA (4-bit quantization + LoRA adapters via PEFT/TRL) on the DialogSum conversational summarisation dataset. Achieved a +5% gain in BERTScore F1 (0.8381 → 0.8805) — demonstrating resource-efficient LLM specialization.

PyTorch QLoRA PEFT / TRL HuggingFace BERTScore

n8n AI Agent & GCP Vertex RAG

Designed and deployed a multi-agent password reset workflow in n8n with OTP generation and token auth. Separately, built a Gemini-based RAG agent on GCP Vertex AI with vector search — significantly reducing LLM hallucinations.

n8n GCP Vertex AI Gemini RAG

Insurance Claim Prediction

ML web application predicting insurance claim approvals using Decision Trees and Random Forest. Achieved 89% accuracy — a 20% improvement over baseline predictions. Visualized with Matplotlib.

Random Forest Pandas Matplotlib

Real-Time Chat Application

Live messaging platform with bidirectional WebSocket communication via Socket.IO. Flask backend, dynamic HTML/JS frontend. Message latency consistently under 200ms.

Socket.IO Flask WebSockets Node.js

Handwritten Digit Recognition

CNN-based MNIST digit classifier achieving 97% test accuracy. Used TensorFlow/Keras with Logistic Regression and SVM baselines. Supports real-time input recognition.

TensorFlow Keras CNNs

Sentiment Analysis

NLP model detecting depressive/negative sentiment in social media posts. NLTK preprocessing, Logistic Regression, WordCloud visualizations. 82% accuracy on 10,000+ tweets.

NLTK NLP WordCloud

Sales Analysis & Forecasting

EDA on historical sales data using Pandas, Seaborn, and Matplotlib. Interactive dashboards built in AWS QuickSight. Resulted in a 15% forecast accuracy improvement.

Pandas AWS QuickSight EDA

Hi, It's Nitish