Mohnish Bangaru - AI Engineer | Open to New Opportunities

Boot Sequence

Aug 2025 – Aug 2026

Founding Engineer

Drizz (New York, NY)

Designed and built Fathom, an autonomous computer-use agent that converts natural-language test intents into executable on-device actions across iOS and Android, orchestrated as a stateful LangGraph state machine with planning, tool-use, grounding, and verification nodes.
Curated 71 agent-tools leveraging map for file access and adding nuanced capability to enable the agent to plan, decide and perform mobile device actions in a sub 4-second step.
Crafted a Sub-agent System that effectively orchestrates and aligns the agent to the task on long running tasks while managing context windows.
Developed a comprehensive Agent Evaluation framework that judges the agent on custom metrics of test completion and task adherence, experimenting with over 200 combinations of prompts and tool structures.

Vision-Language Models LangChain Python LangGraph PyTorch GCP

Sep 2023 – May 2025

Graduate Student Analyst

New York University (New York, NY)

Enhanced data search-ability using the RoBERTa model, improving accuracy in identifying user problems from text comments by 80%.
Created and maintained ETL pipelines for University Data Warehouse (5M+ rows), adhering to strict data governance policies and preprocessing malformed data.
Introduced a set of 15+ Tableau dashboards visualizing key performance indicators to support data-driven decision-making across university departments.

Microsoft Excel Tableau Qualtrics Docker Oracle DB sci-kit learn Python

Apr 2022 – May 2023

Data Scientist

KPMG (Bangalore, IN)

Automated reporting workflows using Python scripts, saving 264+ hours annually by extracting financial data from unstructured documents using deep-learning techniques.
Achieved a data extraction accuracy of 99% using BERT-based models for Named Entity Recognition (NER) on financial documents.

Jan 2021 – Mar 2022

Analyst

Revamped SQL queries for real-time data access, cutting execution time by 30%.
Maintained documentation and scripted REST API's serving Forecasting Models deployed on AWS.

Alteryx API Development AWS API Documentation Python

Applications

Clearbook

A personal-finance app for iPhone you can ask about your own spending. The AI runs entirely on the phone using Apple Intelligence, so your financial data never leaves your hands — and it never does the math either. Real code computes every number; the AI just explains it.

Apple Intelligence On-Device LLMs Privacy by Design SwiftUI

A Low-Resource Parameter-Efficient Fine-Tuning Framework

Big AI models are expensive to retrain, so I worked out a way to teach them new tricks on a small budget. Training got about ten times faster and cost nearly half as much.

Parameter-Efficient Fine-Tuning Quantized Low-Rank Adapters Residual Connections Efficient Inference

Efficient Deep Learning for Edge Deployment

Shrank an image-recognition model to half its size so it runs on small, everyday devices — about ten times faster, no big servers needed.

Parameter Reduction Quantization Efficient DNN Training

Medical Expert Question-Answering Model

Taught a model to read chest X-rays and describe what it sees the way a doctor would. It names the right condition 96% of the time.

Multimodal Fine-Tuning LLMOps Vision-Language Models

Fine-Tuning LLMs using QLoRA

Spent a while teaching language models to follow instructions on a shoestring, testing setting after setting to find the one that actually works best.

Fine-Tuning LLMs Experimentation Deep Learning

Training Data

Sep 2023 – May 2025

Master of Science in Computer Engineering

New York University

Jul 2017 – May 2021

B.Tech in Computer Science and Engineering

SRM Institute of Science and Technology

Ping Me!

Say hi, pitch me something fun to build, or ask about anything on this desktop. I actually reply.