Mownetharan

I build the systems that make models useful.

Applied AI engineer, IITM alum. I build agents, retrieval and voice systems, and the tools that make the work easier.

Bengaluru, India · open to interesting problems

Work you can explore.Built to be useful.

From a benefits screener in production to small, focused open-source tools. Here's what I've been building.

Health Eligible

A conversational benefits screener helping Rhode Island households check eligibility for food, health, cash and child care assistance.

Explore Health Eligible

Open-source tools

All on GitHub

Applied AI / Python

ML-Agent

AutoML with a boundary around the agent. An LLM proposes experiments; ordinary code validates, runs and selects them.

Reading a million cellsMixed values · standalone streaming API

Elapsed time lower is better

openpyxl4.921 s

rustpyxl0.489 s

Peak process memory

openpyxl47.1 MiB

rustpyxl32.8 MiB

Document processing / Rust + Python

rustpyxl

A streaming Excel engine that reads and writes rows in bounded batches, with an opt-in openpyxl compatibility bridge.

About this benchmark

100,000 rows × 10 columns, mixed values. Median of three fresh-process runs on macOS arm64, measured 10 October 2026 with CPython 3.14.5, openpyxl 3.1.5 and rustpyxl 0.1.0.

Both readers use streaming modes. Timings include reading and Python value conversion; peak memory includes the whole process. These results cover the standalone API. The compatibility bridge has separate measurements; results vary by file and hardware.

pyinterpret

One Python API for local and global model explanations.

Model interpretation / Python

ERP-Assistant

Natural-language operations and org charts for ERP workflows.

AI workflows / Python + LangGraph

jsonpath-rx

Query, inspect and update JSON with Rust and Python.

Structured data / Rust + Python

pydocx-fast

Drop-in performance fixes for python-docx, with byte-identical output.

Document processing / Python

See timings · 50,000 paragraphs

Write + save lower is better

python-docx17.6 s

pydocx-fast5.5 s

Load + read all text

python-docx1.35 s

pydocx-fast0.41 s

Published project timings. The test uses separate processes and checks matching output. Hardware and peak memory were not recorded in the report.

Benchmark and test

MownetharanBengaluru, India

From aerospaceto applied AI.

I'm Mownetharan, an applied AI engineer based in Bengaluru. I studied aerospace engineering at IIT Madras, with elective work in data science.

My work spans conversational and voice agents, retrieval, and evaluation. I also build open-source Python and Rust tools for model explanations and document processing.

Download résumé PDF · 96 KB

Where I've worked.What I've worked on.

Applied AI, analytics, and the engineering around them.

ITO Health, PBC→

Health Eligible helps Rhode Island households check eligibility for SNAP, Medicaid, RI Works and child care assistance in one conversation.

roleApplied AI engineer

sinceDec 2025

Genpact→

Banking & capital markets, analytics embedment. Generative AI and analytical solutions built into the analytics org, and the LLMOps that keeps them running once they are.

roleAM, generative AI

whenDec 2024 – Apr 2026

Indivia AI→

Built the backend for personalized marketing emails and product recommendations.

rolefounding AI engineer

whenAug – Dec 2025

IQVIA→

Advanced analytics.

roleanalyst, advanced analytics

whenJul 2022 – Dec 2024

Something in mind?Let's build it.

I help businesses build AI agents: conversational and voice systems, workflows with tools and retrieval, and the evaluation to see what works before shipping.

mownetharan@gmail.com

Download résumé PDF · 96 KB

Prefer a conversation? Pick a time on the calendar.

Book a chat cal.com/mowne/chat

Book a chat on cal.com ↗