AI / ML Engineer

Div Agarwal

Building AI systems at the intersection of research and product. NYU CS · Full-stack execution.

I build AI systems that have to survive both research constraints and production reality: retrieval pipelines, ML workflows, and full-stack interfaces that move from prototype to shipped product. Currently, I'm working on ML-based risk prediction and data pipelines at Daylit, and building agentic research systems and LLM-driven agents for financial analysis at Variant Avatars.

I care about the gap between a model that works in a notebook and a system that holds up when real users touch it. Most of my projects live in that gap: multi-agent pipelines for financial reasoning, serverless automation workflows, and full-stack products with real interfaces on top of AI.

Previously: Software Engineer at AdsGency AI (San Francisco). Researcher at NYU Neuroinformatics Lab.

Experience

AI Engineer (Contract)—DaylitApril 2026 – Present · San Francisco, CA

•Improved A/R collections signaling using ML-based risk prediction models to prioritize and triage accounts.
•Built data pipelines and feature engineering workflows with real-time ingestion and orchestration via Apache Airflow.
•Enhanced conversation summarization across invoices, call transcripts, and notes to improve collections context and decision-making.

AI Engineer (Contract)—Variant AvatarsMarch 2026 – Present · San Francisco, CA

•Improved agentic research system for data retrieval and ranking, increasing result relevance by 30%.
•Built LLM-driven agents for portfolio analysis, enabling automated market data interpretation.
•Developed multi-agent pipelines for financial reasoning and multimodal content generation (audio, presentations).

Software Engineer—AdsGency AIFeb 2026 – March 2026 · San Francisco, CA

•Built a natural language querying system generating SQL via DuckDB across multiple data sources.
•Developed a ranking and scoring system optimizing data selection for audio, image, and video ad generation.

Researcher—NYU Neuroinformatics LabOct 2024 – Feb 2026 · New York, NY

•Trained ML/DL models for neural event detection — achieving 0.92 F1 on voltage imaging data.
•Built a browser-based tool for interactive exploration of calcium trace data used by the lab team.

Data Analyst—AdeptmindFeb 2023 – Jul 2023 · Toronto, Canada

•Scaled landing pages from 10K to 80K, driving a 20% monthly traffic increase via LLM-generated content and SEO.
•Built a Streamlit dashboard for monitoring performance across 19 clients.

Research Engineer—EnordNov 2022 – Feb 2023 · New Delhi, India

•Worked on state of the art depth estimation models using Deep Learning for Autonomous Drone Navigation with ROS and Nvidia systems.

Head of Autonomous Department—Team OJAS (Formula Student)Sep 2020 – Feb 2022 · VIT Vellore, India

•Led team to become the first Indian qualifiers for Formula Student Hungary.
•Implemented YOLO-based detection achieving 92% accuracy; managed perception, planning, and control tracks.

Education

New York University2023 – 2025

MS in Computer Science · GPA: 3.76 / 4.0

Machine Learning, Deep Learning, Computer Vision, NLP, Big Data, Cloud Computing

VIT University, Vellore2019 – 2023

BTech in Computer Science and Engineering · GPA: 8.4 / 10

AI/ML/DL, Statistics, Finance, Business Management

Delhi Public School, R.K. Puram2017 – 2019

High School — Science and Math · GPA: 92.8%

Skills

Core: Python, PyTorch, SQL, React / Next.js, AWS, Docker, Git

Also: HuggingFace, LlamaIndex, RAG, LangGraph, FastAPI, TensorFlow, Pandas, NumPy, Solidity, ROS, MATLAB

Projects

All projects with detail →

Neuro Window Explorer
Browser-based calcium trace analysis tool built at NYU Neuroinformatics Lab. Surrounding research workflow reached 0.92 F1 event-detection score.
Next.js, TypeScript, Python, ML
Expert-Call RAG Assistant
Local RAG stack with Llama 3.1, LlamaIndex, and LanceDB for source-grounded answers over expert-call transcript archives.
Python, Gradio, LanceDB, LlamaIndex
Cloud-Native Social Media Automation
Serverless AWS pipeline connecting image retrieval, Rekognition analysis, Claude captioning, and OpenSearch into one automated publishing flow.
AWS Lambda, Bedrock, OpenSearch, Python
Collections Strategy Management System
Full-stack AI product for debt collections: FastAPI + React + OpenAI, Excel/PDF ingestion, and AI-generated multi-step strategy timelines.
Python, FastAPI, React, OpenAI API

Other experiments

Concurrent Multi-User RAG System · Fine-Tuning Llama 3.1-8B (LoRA, 0.85 accuracy) · Scalable Vector Search Music Recommendations (Spotify dataset) · Trade Statistics Visualization (OECD, Streamlit) · ERC-721 NFT Smart Contract (Solidity, IPFS) · GPT-4 Pokémon Showdown Agent

Personal

Outside of work I take long walks, attend orchestral concerts, and cook. I'm drawn to classic fiction and philosophical books. My music taste runs from Dire Straits and Pink Floyd to Steven Wilson and Pearl Jam.

Contact

Drop me a message — I'll get back to you as soon as I can.

Your information is safe and will never be shared with third parties.

Book a call

Pick a slot and let's talk.