$ whoami

BCA student specializing in Data Science at Sri Balaji University, Pune,
building toward Data Analytics and AI Evaluation. I turn messy datasets
into tested pipelines, interpretable models, and dashboards — then explain
what the evidence can and cannot prove.

Alongside coursework, I interned as a Cloud Application Developer at
Codefirst Technology. My work lives after the model: data quality,
evaluation, business trade-offs, and human review.

$ cat .profile

ROLE        =  Data Analyst  |  AI Evaluation
STATUS      =  BCA (Data Science) Student — Sri Balaji University
DOMAIN      =  Analytics  |  Data Quality  |  AI Evaluation
TOOLS       =  Python  |  SQL  |  Power BI  |  Excel  |  R
INTERNSHIP  =  Cloud Application Developer — Codefirst Technology
PORTFOLIO   =  TrainLens  |  Customer Churn Analysis  |  LLM Safety Eval Benchmark
LOCATION    =  Pune, India
OPEN_TO     =  Data Analyst  |  AI Trainer  |  AI Evaluation Roles

Customer-support data quality and evaluation platform: five quality dimensions, eleven checks, Groq batch labeling with rule-based fallback, confidence-based human review queue, and accuracy / F1 / calibration reporting. Benchmarked on 852 synthetic conversations at a 98.85% quality score.

Stack: Python · pandas · DuckDB · Streamlit · Plotly · scikit-learn Live demo: TrainLens Dashboard · Repo: Nilesh-builds/trainlens

A controlled benchmark scoring AI responses across 9 dimensions — instruction following, factuality, relevance, bias, toxicity, refusal quality, prompt injection resistance, hallucination, consistency — entirely on free-tier APIs. Rule-based checks plus a 2-model LLM-judge ensemble with 95% bootstrap intervals. Judges validated against references (11/11) and blind human review (human-vs-human κ=0.902). GPT-OSS-120B composite 4.28 vs GPT-OSS-20B 4.18.

Stack: Python · Groq free-tier APIs · pandas · matplotlib · Streamlit · Jupyter Live demo: LLM Evaluation Evidence Dashboard · Repo: Nilesh-builds/llm-safety-eval-benchmark

Production-style churn analysis: data-quality checks, SQL views, leakage-safe modeling with cross-validation and calibration, cost-sensitive thresholds, and a live human-review dashboard. Balanced Random Forest chosen on business reasoning, not accuracy alone.

Stack: Python · SQL · Pandas · scikit-learn · Streamlit Live demo: Customer Churn Decision Support · Repo: Nilesh-builds/customer-churn-analysis

Six automation workflows streamlining HR end-to-end: employee onboarding, leave management, sentiment & feedback analysis, policy Q&A bot, AI resume screener & ranker, and a WhatsApp HR chatbot — Google Sheets as the shared data store, GPT-4 as the AI layer, Gmail/Slack/WhatsApp for alerts.

Stack: n8n · Google Sheets · OpenAI GPT-4 · Gmail/Slack/WhatsApp Repo: Nilesh-builds/ai-hr-automation-suite

$ ls /tech-stack --grouped

languages/   python  r  html  css  bash
data/        postgres  mysql
cloud/       aws  git  github  vscode

Cloud Application Developer (Intern) — Codefirst Technology

Hands-on experience building cloud-native applications, applying AWS fundamentals alongside coursework in data science.

AWS Cloud-Native Architecture Application Development

learning:
  - Advanced machine learning & model evaluation
  - Power BI dashboard design for business storytelling

building:
  - trainlens                   # Customer-support data quality, AI labeling, and evaluation dashboard
  - llm-safety-eval-benchmark  # 9-dimension LLM safety benchmark, free-tier APIs, judge validation
  - customer-churn-analysis   # Telco churn EDA + ML + Power BI
  - ai-hr-automation-suite    # 6 n8n workflows automating HR processes

studying:
  - BCA, Data Science — Sri Balaji University, Pune

open_to:
  - Data Analyst roles
  - Business Analyst roles
  - AI Trainer roles

_{// student by day  |  building an analytics portfolio, one dataset at a time}