Utkarsh Sonawane·B.S. in Computer Science · Penn State University

I build software I actually want to use every day.

I’m an honors Computer Science student at Penn State. Most of my projects start with something I want to improve in my own life — my nutrition and training, my setup, or something repetitive I wish worked better. Right now I’m especially interested in LLMs, agentic systems, and backend infrastructure.

Selected WorkView Résumé (PDF)GitHub
Flagship System
1,041 Automated Tests · Deterministic Decimal Math

Nytr: Evidence-Driven Nutrition & Training Decision Platform

Separating authoritative source data, deterministic computation, and user-approved policy changes. Integrates Penn State dining menus, Apple HealthKit telemetry, and Hevy workout revisions via a FastAPI modular monolith, Supabase PostgreSQL, and a native SwiftUI companion client.

PythonFastAPIPostgreSQLSwiftUIHealthKitDocker
Read Technical Case Study

Projects

Selected Work

Flagship System01

Nytr

An evidence-driven personal nutrition and training decision platform integrating campus dining, HealthKit, and Hevy

Technical Case Study

A production-tested personal decision system built on a FastAPI modular monolith, PostgreSQL/Supabase, and a native SwiftUI/HealthKit iOS companion. Separates authoritative source data, deterministic computation, and user-approved policy changes through strict source authority, immutable versioned evidence, replay-safe mobile sync, PostgreSQL RLS, and deterministic Decimal calculations.

Solo Creator · Systems Architecture, Backend, Database & Native iOS
Core Philosophy: DATA → CALCULATION → DECISION
Backend Tests: 1,041 automated tests
Sync Engine: Dual-anchor idempotent sync
Precision: Deterministic Decimal arithmetic
PythonFastAPIPostgreSQLSupabaseSwiftUIHealthKitHevy APIDockerPytest
Read Deep Dive
Core Algorithm02

Clage

From-scratch NEAT neuroevolution implementation and 2D artificial-life simulation

Technical Case Study

Complete NeuroEvolution of Augmenting Topologies (NEAT) implemented from first principles in pure Python without neat-python or external ML libraries. Features a global innovation ledger, topological crossover, structural mutation, speciation, and directed graph feed-forward network execution. Evaluated on classic benchmarks (OR, AND, XOR, sine) across fixed seeds and deployed in an energy-constrained 2D artificial life world.

Solo Creator · Evolutionary Algorithms, Simulation & Analytics
Passing Tests: 188 passing tests
Dependencies: Zero ML packages (pure Python)
Validation: Multi-seed on OR / AND / XOR / Sin
PythonAlgorithms from ScratchComplex SystemsSimulationPytest
Read Deep Dive Repository
03

Devvy

Zero-dependency local daemon coordinating Discord Rich Presence across developer environments

Repository

Local background infrastructure that speaks Discord's local IPC protocol directly over Unix domain sockets with custom binary opcode framing, handshake negotiation, and backoff reconnects. Unifies activity status across OpenCode, Command Code, and VS Code with priority arbitration, heartbeats, and TTL-based source expiry.

Solo Creator · Systems Programming, IPC & Devtools
Runtime Deps: Zero external dependencies
Protocol: Direct IPC binary opcode framing
Validation: 22-scenario automated suite
TypeScriptNode.jsDiscord IPCmacOS launchdUnix Sockets
Repository
Hackathon Finalist04

ScholarAI

AI platform for discovering and verifying Indian government schemes across 1,000+ programs

Repository

Built for the USAII Global AI Hackathon 2026 (6,081+ participants worldwide, 5-person team) — Finalist. As the Data Engineer, I built the complete data engineering pipeline: scraped and normalized 1,000+ complex schemes via regex-based extraction into structured relational models, generated high-dimensional semantic embeddings, and built batched loaders into PostgreSQL with pgvector for hybrid retrieval.

Data Engineer · Data Pipeline, Relational Schema & Vector Database
Hackathon Result: Finalist (6,081+ participants)
Data Volume: 1,000+ schemes indexed
Engineering Team: 5-person team
PythonPostgreSQLpgvectorFastAPISeleniumRAG
Repository

Additional & Exploratory Projects

MedClarity

Client-side audited medical document intelligence with page-level ground truth

Schema-enforced Gemini pipeline comparing cross-document medical reports, identifying contradictions, unit mismatches, and temporal trends. Client-side PDF extraction guarantees that only relevant clinical text reaches the model, paired with page-level clickable citations.

ReactTypeScriptExpressGemini API
RepositoryLive Demo

ESP32 Soil Moisture Monitor

Embedded capacitive soil-moisture sensor with Wi-Fi alert daemon

Low-power embedded telemetry system using an ESP32-C3 microcontroller, calibrated capacitive soil sensing, running window ADC noise filtering, and automated SMTP/Gmail alert dispatching.

ESP32-C3C++ArduinoEmbedded Systems
Repository

About

Why I build

I started building things because I wanted to improve parts of my own day-to-day life. I like making software that becomes part of my routine instead of something I build once and never use again. Nytr is a good example. I train regularly, and I wanted something that could scan the dining options at Penn State each day and recommend meals that fit my nutrition goals. So I built it. I genuinely use Nytr every day to help make decisions about my nutrition and training.

Not everything I build starts with a practical problem, though. Sometimes I build something simply because I want to understand how it works. When I started tinkering with autonomous systems, I became interested in how behavior could emerge from the algorithms underneath them. That led me to neuroevolution and eventually to Clage, where I implemented NEAT myself instead of treating it as a library call. That’s usually how I learn: build something, understand it from the inside, and keep improving it as I learn more.

Right now, I’m spending most of my time exploring LLMs, agentic systems, and backend infrastructure.

Recognition & Leadership

Experience, Honors & Technical Leadership

Leadership & Community

CS Team Lead & Technical Team MemberSept 2025 — Present

IEEE Student Branch · Penn State Harrisburg

Conducted hands-on developer workshops for 20+ engineering students on documentation, Markdown tooling, and developer workflows. Coordinate technical team initiatives and collaborative sessions.

Founder & Editor-in-ChiefJun 2024 — Feb 2025

Glimpse — Student Publication

Scaled technical publication readership from 0 to 250+ subscribers within 3 months. Authored and edited 30+ long-form analytical articles on technology, software, and student engineering.

Honors & Additional Learning

USAII Global AI Hackathon 2026 — Finalist

Selected as finalist among 6,081+ global participants across all tracks (ScholarAI data engineering pipeline).

Dean’s List & Capital Honors Program

The Pennsylvania State University · B.S. in Computer Science (GPA: 3.53 / 4.0).

Additional Learning

Completed: Ethics of AI (University of Helsinki) · AI Essentials (Penn State)

In Progress: Machine Learning Specialization (DeepLearning.AI) · Practical Deep Learning for Coders (fast.ai)

Technical Skills

Tools I use

Languages

PythonTypeScriptJavaScriptC++SwiftSQL

Backend & Systems

FastAPINode.jsExpressFlaskREST APIsDockerLinuxGit

Data & Databases

PostgreSQLSupabasepgvectorVector EmbeddingsRAGNumPyPandas

Mobile & Frontend

SwiftUIHealthKitReactNext.jsTailwind CSS

Embedded & Hardware

ESP32-C3ArduinoSensor Telemetry

Contact

Get in touch.

I’m always open to discussing LLMs, agentic systems, backend infrastructure, or undergraduate research. Feel free to reach out.

utkarshsonawane@psu.eduDownload Résumé (PDF)

You can also reach me directly at utkarshsonawane67@gmail.com

LinkedIn GitHub Harrisburg, PA