OctagonIt

Research and case studies

We take on research projects in AI, data and software engineering, and turn promising results into working products.

Propose a research project See case studies
Research workspace with data visualisations and handwritten notes

Research areas

Where we explore, experiment and build proofs of concept.

Applied machine learning

Prediction, classification and optimisation models tested on real business data.

Computer vision

Image and video analysis, detection and visual inspection.

Natural language processing

Text understanding, search, summarisation and conversational AI, including local-language data.

Data science & analytics

Exploratory analysis, statistics and dashboards that answer specific questions.

Software engineering

Architecture, performance and the design of scalable, maintainable systems.

Emerging technology

Early experiments with new tools and techniques before they reach production.

How we work on research

Industry R&D

Feasibility studies and proofs of concept that test an idea before you invest in a full product.

Research collaboration

Joint projects with universities, labs and research teams on applied AI and software problems.

Prototype to product

When research works, we turn the prototype into reliable, production-ready software.

Our research process

  1. 01

    Define the question

    We agree on what we want to learn and how success will be measured.

  2. 02

    Review & collect data

    We study existing work and gather or prepare the data the project needs.

  3. 03

    Experiment

    We build prototypes and models and test different approaches.

  4. 04

    Evaluate

    Results are measured against the agreed metrics, with honest conclusions.

  5. 05

    Report & hand over

    You receive a clear report, code and next-step recommendations.

Published research

Work our team has researched, written up and published.

Our first published research

Integrated Predictive Intelligence Tool for Pediatric Health (PediTrack)

An AI-powered mobile decision support system for children aged 2 to 6. It combines growth and risk prediction, personalised nutrition, real-time emergency response and a multilingual voice assistant in one platform for caregivers.

Research team of 4, including our founder, guided by 2 supervisors

  • Healthcare AI
  • Machine learning
  • Mobile health

Case studies

Real problems, our approach and the results.

Healthcare & applied AI

PediTrack: AI health platform for young children

Final-year research project at SLIIT, 4-member team

Challenge
Children's health tools look at growth, nutrition and behaviour separately, emergency help is often slow, and many caregivers struggle with apps that only work in English.
Approach
A React Native app on a microservices backend with four AI components: growth and risk prediction (LSTM-DNN), personalised nutrition with a food image scanner, emergency risk scoring with video consultation, and a voice assistant.
Result
A working mobile system with a knowledge base of 24,855 medical documents for its assistant, and voice support in Sinhala, Tamil and English.
  • AI & machine learning
  • Mobile apps
  • Research
View on GitHub
Automotive services

Service centre management system

Built for Neo Tech Motors and Services

Challenge
HR, finance, inventory, appointments and vehicle records were handled through separate manual processes, and customers had no way to book or check anything online.
Approach
A MERN web platform with a customer portal and a staff portal covering eight modules, from employees and finance to appointments and mobile services. We later ran a full security review and hardened the system.
Result
Eight business functions in one system. The security work fixed critical issues including XSS, CSRF and missing access control, with over 94% test coverage on security-critical code.
  • Custom software
  • Security
View on GitHub
Food delivery

CraveCart: food ordering and delivery platform

Customer, restaurant, admin and rider apps

Challenge
Connecting customers, restaurants and delivery riders needs ordering, payments, tracking and notifications that keep working as order volume grows.
Approach
Six Node.js microservices for users, restaurants, orders, payments, delivery and notifications behind an API gateway, with React web portals and a React Native rider app, deployed on Kubernetes in Google Cloud.
Result
Three web portals and a mobile rider app, with Stripe payments and email, SMS and push notifications.
  • Cloud
  • Mobile apps
  • Custom software
View on GitHub
Health & AI agents

Multi-agent symptom analysis system

Runs entirely on local, open-source AI models

Challenge
Symptom journals are free text, so patterns and warning signs are easy to miss, and health data is too sensitive to send to cloud AI services.
Approach
Four AI agents coordinated with LangGraph: one extracts symptoms, one finds patterns and compares them with weather data, one assesses risk and one writes the report. Models run locally through Ollama, with a FastAPI backend and a React frontend.
Result
A private pipeline that turns a journal into a structured health report with risk levels, without the data leaving the computer.
  • AI & machine learning
  • LLM agents
View on GitHub
Environmental data

Coral reef bleaching risk predictor

Trained on the Global Coral Bleaching Database, 1980–2020

Challenge
Bleaching risk depends on many ocean measurements at once, such as temperature anomalies, heat stress, turbidity and depth.
Approach
Four models (logistic regression, random forest, XGBoost and SVM) vote on each prediction, served through a web app with single-site, sample-site and batch predictions.
Result
Random forest reached 90.5% test accuracy with a ROC-AUC of 0.97, and the four-model vote gives a five-level risk rating.
  • Machine learning
  • Data
View on GitHub
Automotive services

AutoCare Pro: cloud-based service centre system

Team of 4, deployed on AWS

Challenge
Bookings, mechanic jobs and payments at a vehicle service centre have to work together, while each part can still be changed on its own.
Approach
Four Node.js microservices for customers and vehicles, appointments, jobs and payments behind an Nginx gateway, with a React frontend, Docker and automated tests in Jest.
Result
A complete booking-to-invoice flow: finishing a job automatically updates the customer, appointment and payment services.
  • Cloud
  • Custom software
View on GitHub