// DATA SCIENCE × AI — RUTGERS '27

THE INCREDIBLE RONIL BASU ISSUE #27

I'm Ronil. I study data science at Rutgers and build AI agents, prediction engines and tools people actually use. A few of them are below with live stats straight from GitHub.

See My Work GitHub ↗
BUILD · SHIP · REPEAT
SCROLL
3 AI ROLES · 2 HACKATHON AWARDS

01 — WHO IS RONIL

DATA SCIENCE STUDENT TURNED AI BUILDER

I'm a senior at Rutgers studying data science with a statistics minor, graduating May 2027. Right now I grade model outputs for a frontier AI lab through Mercor and study AI engineering with CodePath. Before that I prototyped LLM compliance tools as an applied AI intern at Certa AI.

11
Public Repos
3
AI Roles
2
Hackathon Awards
3
Years Building

02 — WHAT I'VE BUILT

HOT OFF THE PRESS

ISSUE #01
2ND PLACE · INTUIT HQ

Intuit SMB Underwriting

An AI that decides which small-business loans are safe to fund. Took 2nd place at Intuit's hackathon.

Pythonscikit-learnSciPy

—

ISSUE #02
BEST USE OF ELEVENLABS · HACKUSF 2026

StormLink

Co-built a disaster-response app that turns live storm data into voice alerts for people in the path.

Google ADKElevenLabsNOAA/FEMA

live demo ↑

ISSUE #03

Marginalia

Turns class PDFs into handwriting-ready notes, each one pinned to the exact sentence it came from. Runs fully offline on a local model.

FastAPIReactOllama

—

ISSUE #04

Structured Concurrency

Cleans up after AI coding tools by killing dead processes before they pile up and slow your computer.

PowerShellBashLinux

—

ISSUE #05

Courtside

Crunches NBA stats to find player bets the sportsbooks priced wrong.

PythonXGBoostPandas

—

ISSUE #06

spotify-cleaner

Finds the songs you never play and clears them out of your Spotify library. Dry-run by default, and deleting anything means typing DELETE first.

PythonFastAPIReact

—

03 — WHERE I'VE WORKED

ON THE CLOCK

01

CodePath AI Engineering Student · Sep 2026 – Present

Ten-week applied AI engineering course, running this fall.

  • Finetuning DistilBERT for multiclass classification on manually labeled text and evaluating it against an LLM baseline given no examples, reporting precision, recall, and F1 per class and analyzing errors with confusion matrices.
  • Building a Flask API that detects text generated by AI, fusing scores from an LLM classifier with stylometric features (variance in sentence length, lexical diversity) into AI/human/uncertain labels.
02

Mercor AI Trainer (Contract) · Jun 2026 – Present

Math and reasoning evaluation for model training.

  • Evaluated model outputs for a frontier AI lab on math reasoning and instruction following, scoring responses against detailed rubrics and ranking pairwise comparisons to produce reliable training signal.
  • Authored structured critiques and reference responses flagging hallucinations, logical gaps, and unmet constraints, improving response accuracy and stepwise reasoning in later training rounds.
  • Designed multistep math and statistics problems to expose failure modes, documenting reproducible cases that fed model evaluation and refinement.
03

Certa AI Applied AI Intern · Jun – Aug 2025

LLM prototypes for vendor compliance review.

  • Prototyped an LLM compliance pipeline to sort 500+ vendor documents and label their clauses by the 5 SOC 2 Trust Services Criteria, then demoed it to managers.
  • Built a RAG drafting demo that prefilled vendor questionnaire answers with context retrieved from prior submissions and policy documents.
  • Benchmarked 6 LLM providers across zero shot, few shot, and chain of thought prompting and helped tune confidence score calibration to improve extraction precision without raising the cost per query.

04 — WHAT I WORK WITH

THE TOOLBELT

Languages

PythonJavaC++TypeScriptJavaScriptGoSQLRRustBashPowerShell

Frameworks & Libraries

FastAPIReactNext.jsNode.jsPyTorchscikit-learnXGBoostLightGBMpandasNumPySciPystatsmodelsLangChainHugging Face TransformersSentence Transformers

Developer Tools & Cloud

GitGitHub ActionsCI/CDDockerLinuxAWSGoogle CloudRailwayPostgreSQLpgvectorSQLiteRedisSupabasepytestREST APIs

Machine Learning & Data

RAGSemantic searchVector databasesAgentic AIMultiagent systemsGoogle ADKA2AMCPLLM finetuningLoRA / PEFTvLLMTRLbitsandbytesPrompt engineeringLLM evaluationETLData pipelinesFeature engineeringModel calibrationTime series validation

Certification Securities Industry Essentials (SIE), FINRA · Sep 2025

05 — LET'S TALK

READY TO BUILD SOMETHING?

I'm looking for internships and interesting problems. The fastest way to reach me:

Email Me GitHub ↗ LinkedIn ↗