Arnav Daultani
New York, NY

Resume

The full version, in text. Project write ups with the tradeoffs and the failures live under work.

arnavdaultani@gmail.com, ad7994@nyu.edu

education

New York University, Tandon School of Engineering

sept 2025 to may 2029, New York, NY

B.S. Computer Science, minors in Data Science and Mathematics

GPA 3.87. Tandon Dean's List, 2025 to 2026.

coursework
Data Structures and Algorithms, Design and Analysis of Algorithms, Object Oriented Programming, Discrete Mathematics, Probability and Statistics, Computer Architecture, Principles of Data Science
clubs
NYU Self Drive, team lead for spring 2026, OSIRIS Lab

Oberoi International School

to jun 2025, Mumbai, India

International Baccalaureate Diploma, 41 out of 45

Higher level: Computer Science, Mathematics AA, and Economics.

experience

AEREM Group, Net Zero Lending

Product management intern, jul 2026 to aug 2026, Mumbai, India

  • Audited the solar EPC loan origination flow across its six stages, and flagged four drop-off and manual handoff points for remediation.
  • Scoped product fixes across intake, underwriting, and disbursal, then prioritized two of them by borrower impact weighed against engineering effort.

EY, Forensic and Integrity Services

AI and forensics intern, may 2026 to jul 2026, Mumbai, India

  • Built and deployed a computer vision pipeline that reads vehicle identification numbers out of insurance claim video, using YOLOv8 for detection and EasyOCR for the read, at 96 percent extraction accuracy for an Indonesian automotive client.
  • Added Qwen3-VL so the same check runs across video, documents, and images, served behind FastAPI with consensus voting between the models. Per claim manual review time fell by roughly 80 percent.
  • Fine tuned the models on client data and shipped the system to production, where it produces audit ready PDF reports.

Navi Mumbai Municipal Corporation

GIS and LiDAR intern, aug 2023, Navi Mumbai, India

  • Analyzed more than 500 properties against NMMC construction code using GIS records and LiDAR surveys, cutting verification time by more than 30 percent.
  • Wrote a Python script that cross referenced GIS data against land tax records, flagging boundary violations across more than 400 properties and saving roughly 45 hours of manual review a week.

projects

Beacon, local mission control for Claude Code agents

typescript, 2026, live

A local browser dashboard for every Claude Code session on a machine, with real control over agents it launches and a cross-project ticket board agents read and write over MCP.

Truvio, measuring how brands surface in LLM answers

typescript, python, 2026, live

A three stage agent pipeline that measures how a brand surfaces across LLM answer engines, running three model APIs behind a single audit engine.

Reading VINs out of insurance claim video

python, 2026, shipped

A computer vision pipeline that pulls vehicle identification numbers out of insurance claim video, shipped to production at EY.

Mapping UC Berkeley

python, 2024, shipped

A Google Maps style campus navigation system, with A* pathfinding and a feature that lets a user exclude zones from a route.

Brain, a small programming language

rust, 2022, live

A compiler built from scratch for a custom language supporting arithmetic and variables: lexing, parsing, semantic analysis, and error reporting.

activities

Student ambassador

New York University, jan 2026 to present

I give campus tours and speak at admissions events, then follow up with prospective students and their families by call, email, and social media. Most of the job is answering the question behind the question someone actually asked.

Pre-College Computer Science Summer Institute

UC Berkeley, jun 2024

Coursework and assessment in recursion, data structures and algorithms, and introductory machine learning. I led a team of three on the campus navigation project.

IoT in Governance: smart feedback systems

White paper, One Young India, aug 2023

A public policy proposal for smart feedback terminals in bus stops and railway stations to capture citizen input where it happens rather than through a separate app nobody installs.

The accessibility design was the substance of it: multilingual support and voice input, benchmarked against systems already in use such as Aarogya Setu.

tools

machine learning and vision

PyTorch, TensorFlow, NumPy, Pandas, Matplotlib, OpenCV, YOLOv8, EasyOCR, Qwen3-VL, LLM APIs: Claude, GPT, Gemini, Model Context Protocol

languages

Python, TypeScript, JavaScript, Java, C, C++, Rust, SQL

web and backend

React, Next.js, FastAPI, Redis, PostgreSQL, REST APIs

tools

Docker, AWS, Git

pdf

This browser will not display an embedded PDF. Open it in a new tab instead.