AI models for thesis and capstone projects in the Philippines.
Machine learning models designed, trained, evaluated, and deployed for undergraduate capstone projects, master's theses, and dissertations. Built to survive a panel's questions, and scoped to a student budget.
Most thesis and capstone projects that involve machine learning fail in the same two places: the data turns out to be unusable in the shape it was collected, and the model works in a notebook but has no story behind why it was chosen. Panels ask about both.
We build the model and the evidence around it. That means the data pipeline, the training, an honest evaluation against a held-out test set, the comparison against simpler baselines that a panel will ask about, and a deployed demo you can actually click through during your defence.
This is proven work, not a side offering. Two of the projects in our portfolio started exactly this way: a neural network that predicts the correct steel connection from structural design inputs, and a model that recommends geotextile materials from site parameters. Both were built for graduate research, and both are live and clickable today.
Scoped to a student budget
Research projects are quoted differently from commercial work. Tell us your budget up front and we'll tell you honestly what's achievable inside it.
Built to defend
Accuracy, precision, recall, confusion matrices, and a comparison against simpler baselines, the numbers a panel asks for, prepared before they ask.
A demo that runs
A deployed web interface your panel can use during the defence, not a notebook you have to narrate from a laptop screen.
You understand your own model
We walk you through every architectural decision so you can answer "why this approach?" in your own words, in front of a panel.
What we build for research projects
Most academic machine learning projects fall into a handful of shapes. A short call is usually enough to work out which one yours is, and whether the data you have can actually support it.
Prediction from tabular data
An artificial neural network or gradient-boosted model trained on historical records to predict an outcome from measured inputs. The most common shape for engineering, agriculture, health, and business research, and what both of our portfolio projects are.
Classification
Sorting inputs into categories: material selection, risk banding, diagnosis support, species or defect identification. Comes with a full confusion matrix and per-class metrics.
Computer vision
Image-based models for detection, counting, grading, and quality inspection. Includes guidance on how much labelled imagery you actually need before this is viable.
Natural language processing
Sentiment analysis, topic modelling, and text classification, including on Filipino and Taglish corpora, which most off-the-shelf tooling handles poorly.
Forecasting
Time-series models for demand, traffic, yield, water level, or consumption, evaluated with proper time-aware validation rather than a random split.
Deployment for defence
Wrapping any of the above in a clean web interface with a public URL, so your panel interacts with the model instead of reading about it.
What data you need before we start
This is the question worth answering earliest, because it determines whether the project is possible at all, and a lot of students find out too late.
For a predictive or classification model you need historical records where the outcome is already known. A few hundred rows is a workable floor, but the real requirement scales with how many input variables you have: more variables need more rows. For computer vision you need labelled images, and the honest minimum is higher than most students expect.
If your data is currently in scanned PDFs, inconsistent spreadsheets, or handwritten survey forms, that's normal and it's part of the work rather than a reason not to start. What genuinely blocks a project is having no outcome variable, or having data where the thing you want to predict was never actually recorded.
Working backwards from your defence date
Academic timelines are fixed in a way commercial ones aren't, so we plan from your defence date rather than from a start date.
A classification or prediction model on data that's already tabulated typically needs three to six weeks end to end: data preparation, training, evaluation, deployment, and a walkthrough so you can explain it. Computer vision projects and anything requiring substantial data cleaning run longer.
The practical advice: come to us before your data collection is finished, not after. A thirty-minute conversation about what to record and how to structure it will save weeks later, and it costs you nothing.
Working alongside your adviser
We build the technical artefact. Your adviser owns the research design, the literature review, the framing, and the academic argument, and we work to their requirements, not around them.
If your adviser has specified a particular algorithm, architecture, or evaluation protocol, we implement that. If they've left the choice open, we'll propose an approach with the reasoning written out so you can take it to them for approval before any work starts.
To be explicit about where the line sits: we build and explain models, and we help you understand your own project well enough to defend it. We don't write your paper for you.
Coverage
Where we work
Kleio is based in Pampanga, and research projects run almost entirely over video calls and shared files. The collaboration that matters is around your data, not around a meeting table. We work with students and researchers across the Philippines and abroad.
Based in
Angeles City, Pampanga
Working with students across
Pampanga
Central Luzon
Bulacan
Tarlac
Nueva Ecija
Zambales
Bataan
Metro Manila
Questions people actually ask
The things people ask on the first call, answered here so you don't have to.
Can you build the AI model for my capstone or thesis project?
Yes. That's a regular part of our work. We handle model design, data preparation, training, evaluation, and deployment, plus a walkthrough so you understand every decision well enough to defend it. Two of the AI projects in our portfolio were built for graduate research: a steel-connection classifier and a geotextile material predictor, both live and clickable today.
How much does an AI thesis or capstone project cost?
Research projects are quoted differently from commercial work and scoped to a student budget. Price depends mostly on the state of your data. A clean tabulated dataset is a much smaller job than four years of scanned survey forms. Tell us your budget on the first call and we'll tell you honestly what's achievable within it, rather than quoting something you can't afford.
How much data do I need for a machine learning thesis?
For a prediction or classification model you need historical records where the outcome is already known. A few hundred rows is a workable floor, but the requirement scales with the number of input variables: more features need more rows. Computer vision needs labelled images, and the honest minimum is higher than most students expect. Talk to us before you finish collecting; a short conversation about structure saves weeks later.
How long before my defence should we start?
Three to six weeks is typical for a classification or prediction model on data that's already tabulated, covering preparation, training, evaluation, deployment, and your walkthrough. Computer vision and heavy data-cleaning projects run longer. Ideally reach out before data collection is finished, so the data gets recorded in a shape a model can actually learn from.
Will I be able to explain the model during my defence?
That's a core part of the engagement, not an afterthought. We walk you through why the architecture was chosen, what the evaluation numbers mean, how it compares against simpler baselines, and where the model fails. A panel will ask all four.
Can you deploy the model so my panel can try it?
Yes, and we recommend it. We wrap the model in a clean web interface with a public URL, so your panel interacts with it live rather than watching you narrate a notebook. Both research projects in our portfolio are deployed this way and still running.
Do you write the thesis paper too?
No. We build and explain the technical artefact (the model, the evaluation, and the deployed demo) and we help you understand it thoroughly enough to write and defend it yourself. The research design, literature review, framing, and writing stay yours and your adviser's.
What if my adviser requires a specific algorithm?
We implement what they've specified. If the choice has been left open, we'll propose an approach with the reasoning written out so you can take it to your adviser for approval before any work begins. Your adviser's requirements lead.
Do you work with students outside Pampanga?
Yes. We're based in Angeles City, Pampanga, but research projects run almost entirely over video calls and shared files, so location has never limited one. We work with students across Metro Manila, Central Luzon, the wider Philippines, and abroad.
A free call with the developer who'd build it. Bring your research question and whatever data you have so far. We'll tell you straight whether it's feasible and what it would take.
