94%
Model Recall Rate
Achieved via 1D Convolutional Neural Networks for complex astronomical time series classification.
Available for AI and Data Automation roles
AI Automation Engineer
Fusing a physics mindset with artificial intelligence and scalable data automation.
My engineering philosophy begins where my physics education ended, at the fundamental principles. I break down complex, noisy systems to first principles to build clean, scalable, production-ready AI solutions.
I don't just apply models. I architect them to solve real-world problems.
/ Engineering impact
Outcomes measured in production, not slides.
250,000+
Production Records Processed
Handled via automated data ingestion pipelines and modular Python scripts.
50%
Efficiency Increase
Delivered by engineering robust GUI applications and automated data cleaning workflows.
£0
Cloud Costs
Architected an offline RAG pipeline using local Ollama models and vector databases.
/ Work in the field
Engineering clarity out of complex bottlenecks.
Case Study 1 / Data Automation
Modular Automation for Production Data
40%
faster analysis
The Problem
Researchers at Paragraf were drowning in complex, large-scale experimental spectral datasets, spending dozens of manual hours cleaning and analyzing 250,000+ records.
The Objective
Design a robust, intuitive automated tool to instantly accelerate data insights and speed up technical reporting.
My Action
Engineered a modular Python-based desktop application (GUI) for spectral analysis. Orchestrated modular Python scripts to automate data ingestion and formulated rigorous data cleaning processes.
Resulting in
Reduced manual analysis time by 40%, delivering results 2x faster and boosting raw data operational efficiency by 50% across 250,000+ records for stakeholders.
Case Study 2 / AI Security Tooling
AI Security Scanner CLI
CLI
automated triage
The Problem
Manually reviewing codebases for security vulnerabilities is time consuming and prone to human error. Developers often miss critical flaws like SQL injections or hardcoded credentials during rapid development cycles.
The Objective
Architect an automated command line tool to process local Python files and stream them directly to a Large Language Model for comprehensive security analysis.
My Action
Engineered a Python application integrating the Gemini 2.5 Flash API. Applied advanced prompt engineering techniques to force structured data responses rather than conversational text. Implemented visual triage parsing to automatically colour code vulnerabilities by severity.
Resulting in
An actionable automated security pipeline that successfully identifies critical vulnerabilities and prioritises developer workflows at the command line level.
Case Study 3 / Local AI Infrastructure
Local AI Environment & Custom Personas
0$
offline inference
The Problem
Cloud-based AI APIs introduce privacy risks, ongoing costs, and require constant internet connectivity, making them unsuitable for handling sensitive or highly private local data.
The Objective
Establish a completely secure, zero-cost local Artificial Intelligence infrastructure to run Large Language Models directly on local hardware.
My Action
Deployed a local AI environment using Ollama. Downloaded and configured specific models (Qwen, Gemma) and engineered custom Modelfiles to enforce strict persona behaviours and structured outputs through advanced system prompts.
Resulting in
A fully operational, private AI environment capable of secure offline inference and highly customised, instant model interactions.
Case Study 4 / RAG & APIs
RAG REST API with FastAPI
RAG
grounded responses
The Problem
Large Language Models lack context of private data and often confidently hallucinate answers when queried about specific, personal, or proprietary information outside their training scope.
The Objective
Architect a complete Retrieval-Augmented Generation (RAG) pipeline to ground AI responses in custom documents and serve it through a production-ready REST API.
My Action
Engineered a FastAPI backend connected to a ChromaDB vector database. Developed data ingestion scripts to chunk and embed text documents, and built multi-user GET and POST endpoints using Pydantic for strict data validation and metadata filtering.
Resulting in
A scalable API that accurately retrieves contextual data via semantic search, augments the prompt, and generates perfectly grounded AI responses without any cloud dependencies.
Case Study 5 / Deep Learning
Pulsar Classification Using 1D CNNs
94%
recall
The Problem
Analyzing high-dimensional astronomical time-series data at scale is manually intensive. Identifying critical celestial objects, such as pulsars, from noise was a bottleneck slowing discovery.
The Objective
Take sole ownership to architect, train, and deploy an automated, reliable deep learning pipeline capable of high-precision pulsar detection.
My Action
Optimized experimental data workflows by engineering a 1D Convolutional Neural Network (CNN) in Python. Evaluated multiple configurations to maximize precision, utilising Scikit-learn for comprehensive validation.
Resulting in
A robust production pipeline achieving 94% recall, completely eliminating manual signal triage and enabling scalable astronomical discoveries.
Case Study 5.1 / Deep Learning Expansion
Multi Class Pulsar Classifier
3 Class
softmax output
The Problem
Binary pulsar classifiers fail to distinguish complex geometric pulse profiles and cannot separate rare multimodal signals from authentic telescope interference.
The Objective
Upgrade a foundational binary classifier into a robust multi class architecture capable of identifying intermittent pulsars while ignoring genuine telescope noise.
My Action
Engineered a unified hybrid dataset by extracting genuine background red noise from target physical telescope recordings. Constructed a master training tensor with strict class imbalance to simulate real world astronomical signal rarity, then retrained a 1D Convolutional Neural Network ending in a three neuron softmax layer to output precise confidence probabilities.
Resulting in
A production ready multi class pipeline that isolates uncatalogued multimodal candidates with calibrated confidence probabilities and generates expert ready visual grids for physical verification.
/ The expertise cloud
Tools I reach for instinctively.
An interwoven stack across modelling, automation, and the human side of shipping AI inside teams.
- Python
- Machine Learning
- FastAPI
- Vector Databases (ChromaDB)
- Local LLMs (Ollama)
- Prompt Engineering
- SQL
- Data Automation
- Semantic Search
- CLI Development
- Gemini API
- 1D CNNs
- Scikit-learn
- Data Analysis & Visualisation
- Technical Presentations
- Stakeholder Communication
/ The person behind the pipeline
Engineering is a team sport. Here is mine.
Academic
Physics, with intent
Graduate (2:1 Class Hons) in BSc Physics with a Year in Industry from the University of East Anglia, Norwich. Year in Industry as a Research Intern at Paragraf, Somersham.
Leadership
500+ members, one community
President of the Latin American Society for three consecutive years. Managing logistics, sponsorships and large-scale events for a 500+ member community.
Off the clock
Authentically me
Outside of work you will find me in structured strength training, exploring regional Italian culinary traditions, or deep-diving into complex non-fiction.
/ Beyond the codebase
Leadership & Community
Engineering is not just about building systems; it is about building the teams and communities around them.
I served as the President of the Latin American Society at the University of East Anglia, leading a vibrant community, orchestrating complex event logistics, managing budgets, and aligning diverse stakeholders. This experience forged my ability to communicate technical ideas clearly and lead initiatives with empathy and structure.
500+
Active Community Members
3 Years
Executive Leadership Tenure
+35
Major Events Orchestrated
/ Let's build something
Looking for someone with the skills to bridge data and real world impact?
I am ready to turn complex data into scalable AI solutions.
LinkedIn
daniel-barranco-aragon
GitHub
DanielBarranco-Phy
Email
daniel.barranco.a@gmail.com
Location
UK / Open to relocate