Senior Software Engineer — Autonomous Systems & AI Agents

I plan the next move —
for tractors, and lately, for agents.

Three years writing the C++ that decides where an autonomous vehicle goes next when the ground under it keeps changing. Now I point that same instinct — search, constraints, deterministic decision-making — at LLM-based agents, building the ones that have to reason the same way every time.

Download résumé See the work →

fig. 01 — search → cost-map → replan live

About

// WP-01

Most of my last three years lived at the intersection of geometry and real-time constraints: A*/Hybrid A* search over live cost-maps, path smoothing, turn and trajectory generation under non-holonomic constraints, and multi-vehicle coordination for autonomous agricultural robots — code that has to decide, in milliseconds, what "safe" means before it means "fast."

These days I'm pointing that same discipline at a different kind of agent: at Arcesium, I build LLM-based agents designed to be deterministic rather than merely capable — same task in, same reasoning path and answer out, every time. I keep the fundamentals sharp with 550+ problems solved on LeetCode, and I'm looking to deepen the theory side of all this with graduate study in robotics.

3+ yrsautonomous systems & AI agents

550+LeetCode solved

$2.5Msaved via in-house web app

Experience & Education

// WP-02

JUN 2026 — CURRENT

Senior Software Engineer, Arcesium

Bangalore, India

Design and build LLM-based agents in Python that produce deterministic, heuristic-driven administrative outputs — prioritizing reliability and precision over open-ended generation.
Own the full agent development lifecycle: architecture, structured instruction/skill design, testing, and refinement — defining how an agent reasons through a task and how it presents the final answer.
Build evaluation and testing frameworks to validate agent correctness, consistency, and determinism across edge cases before deployment.

JUL 2023 — MAY 2026

Software Engineer, John Deere

Pune, India

Designed and implemented the full motion planning pipeline in C++ on embedded systems for autonomous agricultural vehicles in unstructured field environments — A*/Hybrid A* search, cost-map construction, path smoothing, and turn/trajectory generation under non-holonomic motion constraints.
Built real-time obstacle avoidance and dynamic replanning that responded to moving obstacles and changing field conditions, improving operational reliability by 25%.
Developed multi-vehicle coordination logic enabling safe, conflict-free operation of multiple autonomous machines working the same field at once.
Led a React.js + Java microservices web app for fleet/operations management that replaced a third-party licensed tool, saving $2.5M a year, while mentoring junior engineers through the build.
Took a field-efficiency idea from sketch to a filed patent, with measurable impact once shipped.

MAY 2022 — JUL 2022

SDE Intern, SSM Infotech

Surat, India

Built a React Native attendance app (Android + iOS) with geolocation-restricted punch-in and automatic working-hours calculation.
Designed advanced SQL — multi-table joins, nested subqueries, window functions — for real-time attendance analytics.
Shipped a responsive React.js UI, plus secure auth and persistent session handling.

JUL 2019 — MAY 2023

B.Tech, Computer Science & Engineering

Sardar Vallabhbhai National Institute of Technology, Surat · CGPA 7.84/10

Skills

// WP-03

Motion Planning & Robotics

A* / Hybrid A*Cost-map planning Path smoothingTrajectory generation Real-time replanningObstacle avoidance Multi-vehicle coordinationNon-holonomic constraints

AI Agents

LLM agent designDeterministic outputs Structured instructionsAgent evaluation

Languages

C++PythonJava TypeScriptRust

Systems & Backend

MicroservicesApache Kafka MySQLMongoDBRDBMS

Infra & Cloud

DockerKubernetesAWS JenkinsCI/CD

Frontend

React.jsReact Native

Foundations

DSAOODSOLID Design PatternsPerf. Optimization

Projects

// WP-04

Kinodynamic RRT* — quadrotor trajectory planner

A from-scratch C++17 motion planner for a double-integrator quadrotor: a closed-form optimal boundary-value solution (depressed-quartic arrival time) serves as both the steering function and the RRT* cost metric, with choose-parent/rewiring for asymptotic optimality — success rate climbs from 15% to 100% and mean path cost drops from 26.9 to 24.5 as the iteration budget grows, producing dynamically feasible curved 3D trajectories through obstacle fields. Ships with an interactive Dear ImGui/raylib editor with background-thread replanning, a headless OpenCV renderer for CI, 27 GoogleTest unit tests, and a GitHub Actions matrix across GCC/Clang.

C++17 · CMake · Eigen3 · Dear ImGui · raylib · OpenCV · GoogleTest · GitHub Actions

AI-assisted code refactoring platform

A web tool that improves code quality and readability while preserving behavior — detecting smells, cutting redundancy, and enforcing standards across languages. Cut manual refactor time and lifted developer throughput by ~30%.

Python · React.js · GitHub integration

Stock market prediction

A CNN-based platform forecasting short-term price trends from chart-based inputs, trained with TensorFlow/Keras and served through a Streamlit dashboard for real-time, interactive predictions.

Python · TensorFlow · Keras · Streamlit

I plan the next move —
for tractors, and lately, for agents.

About