Ammiel Bowen

Transcripter: Transcript Management and Processing System

An internal FSU IT Services system built for the Office of Admissions. It replaces fragmented, manual transcript processing with a structured workflow. Staff submit transcript records, track processing progress, and manage employee roles, and administrators get an analytics view of submission trends and processing activity.

Key features: role-based access control, a submission form with dynamic validation, a real-time dashboard where staff sort, filter, and manage records, a log of every create, edit, and delete, multi-document support, and admin tools for managing users.

The hard parts were keeping audit logs accurate across every action, handling sensitive information securely, building interfaces staff find intuitive, and designing a flexible role-based system. Next on the list: automated document processing, integration with more university systems, better search and filtering, richer activity monitoring, and reporting tools for administrators.

React, Vite, JavaScript, CSS, Node.js, Express, MongoDB, REST API, GitHub

View the poster (PDF)

Internal FSU tool, not publicly available. The poster's dashboard screenshot is pixelated to protect applicant records. Results are under Experience.

Predicting Seizures with Graph Neural Networks

Undergraduate research with Dr. Yushun Dong in FSU's Department of Computer Science, documented in a 2025 Undergraduate Research Opportunity Program (UROP) poster. It tests whether combining graph neural network (GNN) models improves real-time seizure prediction from EEG.

EEG recordings become functional brain networks, with EEG channels as nodes and edges from Pearson correlation and mutual information. Deep Graph Infomax learns node embeddings, and a GNN-based temporal model extracts features over time. Preliminary results show accuracy in the mid-70% range, with better F1 and AUC scores than single-GNN models.

PyTorch, PyTorch Geometric, Deep Graph Infomax, EEG

View the poster (PDF) · View the code

RSO Funding Assistant

As Chair of SGA's RSO Allocation Committee, I answer funding questions from student organizations every week, and most of the answers are already in the 18-page Financial Manual. This assistant answers those questions from the 2026-27 manual only, and cites the section and page.

The manual's own table of contents drives section-aware chunking, with typo-tolerant heading matching. Chunks and embeddings live in Postgres with pgvector. Claude must answer through a structured tool schema, and code checks that every quoted citation actually appears in the passage it cites; when the manual doesn't cover a question, it says so. It also runs as an MCP server, and ships with an eval set and CI that runs integration tests against a real pgvector database. In progress.

TypeScript, Node.js, Express, Postgres, pgvector, Claude API, OpenAI embeddings, MCP, Vitest, GitHub Actions

View the code

Rate My Professor RAG

A retrieval-augmented assistant for finding professors. Reviews are embedded with OpenAI's text-embedding-3-small and stored in Pinecone; each question retrieves the closest reviews, and gpt-4o-mini streams an answer grounded in them to a Next.js chat interface.

Next.js, OpenAI API, Pinecone, Vercel

Open the live site · View the code

Computational Neuroscience Chatbot

A retrieval-augmented generation (RAG) support chatbot for computational neuroscience students. A Next.js API route sends each question to Amazon Bedrock Knowledge Bases, which retrieves passages from a Pinecone vector index and generates an answer grounded in that source material rather than the model's training data alone.

Next.js, Node.js, Amazon Bedrock, Pinecone, RAG

Watch the demo · View the code

SmartCards

An AI-powered flashcard generator. Paste in your notes and it turns them into study flashcards you can open on any device. Built during the Headstarter AI Software Engineering Fellowship.

Next.js, OpenAI API, Firebase Firestore, Clerk, Stripe

Open the live site · View the code

Inventory Tracker

A web-based inventory tracker with item search, running totals for quantity, items, additions, and deletions, and a recent-activity log. Built during the Headstarter AI Software Engineering Fellowship.

Next.js, Material UI, Firebase Firestore

Open the live site · View the code

Cybersecurity Investigation

An analysis of an open-source dataset examining a UI prompt for a credentials function, written for CodePath's Intermediate Cybersecurity course (CYB 102), which I completed with honors.

View the slides