Project highlight
An AI Masters project that turned into a start-up
I turned a hard question about getting started, staying focused, and returning to difficult work into a voice-first multi-agent system. The research prototype became Accai, the startup I am building around AI accountability.
Accai
startup built from the project
14/14
positive task-initiation responses
8.33/10
average prototype rating
Work
These are the work projects: search and retrieval with Solr, and an Alteryx computer-vision roadmap for PCM waste-can inspection.
A search-led knowledge system that uses Solr for discovery and LLMs for inquiry, turning scattered notes into an inspectable retrieval surface.
What it proves
Shows how I design information systems that make memory operational: indexing, metadata, ranking, and question-answering as one workflow.
A no-code computer-vision roadmap translating Alteryx Intelligence Suite into a practical workflow for PCM waste-can X-ray inspection.
What it proves
Shows how I turn broad AI ambition into usable operational judgment: dataset readiness, specialist review, automation boundaries, and responsible next steps.
Projects
Furhat conversational agents, emotional classification, particle-swarm optimization, and machine translation each get treated as standalone projects with visible mechanics and a clear case for what they prove.
ProjectA prompt-engineered LLM controller for a Furhat robot study, connecting embodied dialogue, experiment traces, and social-alignment results.
Combines dialogue design, Kotlin/Furhat engineering, prompt conditions, and statistical interpretation in one research system.
ProjectAn applied machine-learning project comparing classical and neural models for classifying emotional state from EEG and physiological signals.
Shows experimentation discipline across preprocessing, PCA, model comparison, and F1-score reporting on noisy high-dimensional data.
ProjectA visual explainer for training an artificial neural network with particle swarm optimization instead of ordinary backpropagation.
Shows how I make complex optimization mechanics inspectable: particles, weight vectors, fitness, velocity, and convergence tradeoffs.
ProjectInteractive NLP walkthrough for seq2seq translation and attention.
Shows how encoder states, attention weights, decoding strategy, and BLEU-style evaluation fit together in a translation workflow.