AI/ML Engineer & Software Developer

Building intelligent systems from model to deployment.

I’m a Computer Science student at Kent State University graduating in May 2027. I build applied AI systems across the full stack—from model development and computer vision to LLM infrastructure, backend software, testing, and deployment.

Kent State University · Computer Science · Expected May 2027

View Projects Contact Me

GitHub LinkedIn

Selected work

Systems built across the AI stack.

Selected work spanning computer vision, retrieval-augmented generation, and model internals — plus the earlier and supporting projects behind them.

01 / Computer vision

OrbitalVision

Aerial Building Footprint Segmentation

A PyTorch semantic-segmentation project using a custom TinyUNet encoder-decoder with skip connections to predict building footprints from Massachusetts satellite imagery. The goal wasn’t just a working model — it was understanding how a U-Net actually moves data through an encoder, a decoder, and the skip connections between them, and learning to test each piece in isolation before integrating the full pipeline, rather than building everything at once and debugging afterward.

IoU: 0.5280
Dice: 0.6901

Mean per-image scores on a 10-image held-out test split.
Full-resolution inference is not yet implemented.

Visual comparisonTinyUNet · building footprints

Four-panel OrbitalVision test visualization showing an input satellite image, ground-truth building mask, probability heatmap, and predicted binary mask — Test example / input · ground truth · probability heatmap · predicted mask

02 / Local AI

SourceRecall

Local Retrieval-Augmented Generation

A local RAG system for document ingestion, embedding generation, vector retrieval, source grounding, and local model inference — built to answer questions against a personal document set without sending anything to a hosted API.

Core stack: Python · ChromaDB · Sentence Transformers · Ollama
Focus: Local retrieval-augmented generation

Functioning prototype. Retrieval-quality and latency evaluation have not been formalized yet, and the interface is under active revision.

System flowLocal RAG pipeline

Ingest
Embed
Vector store
Retrieve
Ground & generate

Chroma-backed retrieval grounds each response in the source document set before a local Ollama model generates an answer.

03 / Model internals

Transformer From Scratch

Learning Build

A transformer implemented from first principles — attention, multi-head attention, and transformer blocks — to understand how the architecture actually works rather than only consuming it through high-level APIs.

Learning project; not a production system.

Also building

Product · Experimental

Keepwise

A macOS app I’m building to solve a problem I have as a photographer — finding byte-for-byte duplicate photos in a large library — with the long-term possibility of other photographers using it too. I designed the product direction, but Codex implemented a substantial share of the code, so this is supporting, AI-assisted product work rather than a primary engineering showcase.

126 frontend and Rust tests passing. Duplicate matching was also exercised against a 10,000-file synthetic validation corpus on one arm64 macOS system. Exact copies only — visual similarity and cleanup execution are not implemented.

2024 · Archived

PhotoRank-AI

My first AI/ML project — an early attempt at automated photo-quality ranking. Abandoned, but the starting point for the computer-vision interest that led to OrbitalVision.

Experience & Education

Flagship experience

AI/ML Engineering Intern

Rocky, Kent State University · June–August 2026

Worked remotely with a team building Rocky, Kent State University’s internal AI chat platform intended to serve thousands of students. I implemented backend parsing so the API’s OpenAI-style JSON request format could be transformed into what Ollama needed, and built backend telemetry endpoints that the admin panel now consumes — though I didn’t build the admin panel’s frontend myself. Working inside an existing system with limited production server permissions meant learning new Python and library patterns quickly and designing around real constraints rather than ideal ones.

Team: Remote · Kent State AI/ML
Scope: API request parsing · Backend telemetry
Constraint: Limited production server permissions

Internal systems and source remain private.

B.S. Computer Science

Kent State University · Expected May 2027

Cybersecurity concentration.

Owner & Photographer

Nick’s Captures LLC · 2025–Present

Operate a real-estate and drone photography business, managing clients, aerial imaging, and media delivery.

FAA Part 107

Remote Pilot Certificate

Certified for commercial drone operations and aerial imaging.

About

I’m a Kent State Computer Science student focused on AI/ML and software engineering. I like building complete systems—the model, backend, tests, and interface—so technical ideas become useful, reliable tools.

Photography sharpens how I observe, frame problems, and communicate visually, while space and Earth-observation interests guide me toward meaningful technical work.

Contact

Let’s build something meaningful.

I’m open to AI/ML internships and May 2027 full-time roles involving applied AI, computer vision, backend systems, and product-minded engineering.

Email nickvuletich23@gmail.com