About Me

I turn messy, unstructured documents — PDFs, scans, and legacy files — into clean, structured data that your systems can actually use. If your team is drowning in paperwork, archives, or drawings that no software can read, that's the problem I solve.

The results are measurable: full-stack OCR pipelines achieving >90% accuracy on legacy hand-drawn documents (up from under 10%), Neo4j knowledge graphs that make previously unsearchable archives explorable, and TrOCR fine-tuning for domain-specific handwriting. I'm also a published researcher — primary author of ProS2Vi, peer-reviewed in ScienceDirect.

I work with clients on document OCR & digitization, PDF and data extraction pipelines, unstructured-data-to-structured-database ETL, and ML/LLM integration — delivering each project end-to-end, from model fine-tuning to production deployment.

>90%
OCR Accuracy
160K+
Files Processed
8x
Speed Gain
150+
App Users Shipped

Work With Me

Document OCR & Digitization

Service
  • Turn scans, handwritten records, and legacy paper archives into searchable, machine-readable text.
  • Custom OCR and TrOCR fine-tuning for domain-specific handwriting and low-quality documents — proven >90% accuracy on tough legacy material.
OCR TrOCR Computer Vision

PDF & Data Extraction Pipelines

Service
  • Automated extraction of fields, tables, and metadata from dense PDFs and forms at scale.
  • High-throughput, parallelized pipelines — I've processed 160,000+ unstructured files with an 8x speed gain.
PDF Parsing Python Parallel Processing

Unstructured → Structured ETL

Service
  • Transform raw documents into clean, structured databases and knowledge graphs your systems can query.
  • Neo4j knowledge-graph architectures that make previously unsearchable archives explorable via semantic search.
ETL Pipelines Neo4j Knowledge Graphs SQL

ML / LLM Integration

Service
  • Integrate LLMs, RAG, and agentic workflows into your product or internal tooling — delivered end-to-end.
  • From model fine-tuning to production deployment, with a focus on accuracy, reliability, and data privacy.
LLMs Graph-RAG Fine-tuning LangChain
Let's Talk About Your Project

What I'm Building

LectureLearn

Shipped · ~150 users

Creator & Solo Developer

  • An AI study tool I built and shipped as a personal project, now used by around 150 students.
  • Parses uploaded course documents and auto-marks student answers against provided mark schemes.
  • Generates flashcards and quizzes from source material, with text-to-speech for on-the-go review.
  • Full-stack build covering document ingestion, LLM-backed generation, and a responsive study interface.
LLMs RAG Document Parsing TTS Python Full-Stack

Work Experience

Software Developer

Enterprise Engineering & Design Firm

Mount Pearl, NL · On-site

  • Progressed from part-time contract to permanent full-time (June 2026), leading AI integration across the engineering document stack.
  • Architect an agentic interface over a Neo4j knowledge graph for a large-scale knowledge graph system, enabling semantic exploration of industrial CAD data.
  • Develop and deploy custom AI plugins for enterprise CAD platforms, establishing connectivity with external web-based automation platforms.
  • Own the full architectural lifecycle using AI-assisted tooling (Cursor, Claude Code) to ship secure, production-grade solutions rapidly.

Software Development Intern

  • Engineered an enterprise OCR solution by fine-tuning Microsoft TrOCR on legacy hand-drawn engineering documents, boosting recognition accuracy from <10% to >90%.
  • Delivered full-stack solutions — database, API, and UI components — using C#, SQL, JavaScript, and React.
  • Collaborated in a cross-functional team using Agile and Scrum methodologies to ensure timely, efficient project delivery.
  • Leveraged Azure DevOps for project management, source control, and CI/CD pipeline management to automate testing and deployments.

AI Data Specialist (RLHF & Logic)

Data Annotation

Remote

  • Optimize state-of-the-art Large Language Models via Reinforcement Learning from Human Feedback (RLHF), directly shaping production model behavior.
  • Train models to execute complex agentic tasks and improve multi-step reasoning and logic for Python code generation.
  • Conduct rigorous code quality reviews against established accuracy and correctness benchmarks.

Research Assistant — ML & Data Engineering

Memorial University of Newfoundland — School of Pharmacy

St. John's, NL

  • Architected high-throughput parallelized pipelines processing 160,000+ unstructured protein files, achieving an 8x (800%) reduction in processing latency.
  • Primary author of ProS2Vi, a peer-reviewed computational tool published in ScienceDirect for protein secondary structure visualization with DSSP algorithm integration.
  • Optimized data processing algorithms to further reduce computational overhead by 40%.

Projects & Publications

ProS2Vi

Publication · ScienceDirect

Computational and Structural Biotechnology Journal — February 27th, 2025

ScienceDirect GitHub Repository
  • A SaaS-ready bioinformatics engine designed for secure local execution — eliminating data privacy risks of web-based alternatives.
  • DSSP algorithm integration classifying 8 secondary structure categories (vs. 4 in traditional tools), surpassing standard tooling in resolution.
  • Dual-interface architecture: Flask GUI for interactive use and CLI for batch processing workflows at scale.
  • Data security-first design — fully local execution ensures no sensitive research data leaves the machine.
  • Export pipeline generating publication-ready PDFs and high-resolution images via Jinja2/wkhtmltopdf; supports both PDB and mmCIF formats.
Python Biopython Flask Jinja2 DSSP wkhtmltopdf

PDBpepDS

ML Dataset
GitHub Repository Zenodo
  • Large-scale ML-ready dataset engineered from 160,000+ RCSB protein files, purpose-built as a computational drug design resource.
  • 29 distinct descriptors per peptide: molecular weight, hydrophobic moment, isoelectric point, BLOSUM, FASGAI, Kidera Factors, and Z-scale properties.
  • Four segmented output files enabling targeted research: full dataset, 15aa peptides, <15aa peptides, and unknown residues.
  • Published on Zenodo and GitHub; adopted by researchers in peptide therapeutics and protein structure analysis.
Data Mining RCSB PDB Python Pandas Biopython

Technical Skills

Agentic AI & LLMs

Agentic Workflows Graph-RAG (Neo4j) RLHF Fine-tuning LangChain PyTorch Transformers Large Language Models

Computer Vision & OCR

TrOCR Vision Transformers Computer Vision Optical Character Recognition Legacy Digitization scikit-learn TensorFlow

Engineering Systems

CAD Automation CAD Data Extraction Document Parsing C# (.NET) React Full-Stack Development SQL Server

Infrastructure & Delivery

Azure DevOps CI/CD Docker Parallel Processing ETL Pipelines Git Agile/Scrum Shell Scripting

Get In Touch

Have a document-heavy problem to solve — scans, PDFs, archives, or unstructured data? Tell me about your project and I'll get back to you.