Skip to content
Sachini Dilrangi.

open to software engineering roles · ai-powered products

Sachini
Dilrangi

software engineer.Final-year Software Engineering undergraduate at SLIIT. Most recently Associate Software Engineer at Cloud-eDesign.
↓  SCROLL
01The Brief

Why, not
just what

I’m a final-year Software Engineering undergraduate at SLIIT, and most recently an Associate Software Engineer at Cloud-eDesign, where I built and shipped production web apps for client businesses — some solo, some with a team. I’m now targeting roles at companies building AI-powered software.

I care about understanding why a technical decision was made — not just shipping something that works. That means a real industrial workflow — CI/CD, branch protection, honest testing, honest tradeoffs — over tutorial-style projects.

02The Record

Associate Software Engineer

Cloud-eDesign · Aug 2025 — Apr 2026

Shipped production work across a wider set of client projects than the sites listed below — owning some end-to-end and collaborating on others — spanning Next.js frontends, API routes with HTTP-only cookie auth and Google OAuth, PayHere & Genie payment integrations, AWS S3, and Vercel deployments.

Software Engineer Intern

Cloud-eDesign · Feb 2025 — Aug 2025

Built and shipped features on the same production client codebases in Agile sprints with Git and Jira; promoted to Associate Software Engineer after six months.

Skills & technologies used
Frontend

Next.js 15 (App Router) · TypeScript · Tailwind CSS (incl. v4) · React · Redux · Responsive design · TipTap & TinyMCE · HTML-to-JSX converter · SVG filters & animation · Accessibility (WCAG)

Backend & Auth

Next.js API routes · JWT Auth · HTTP-only cookie auth · Google OAuth · Flask (Python) with OCR · Spring Boot

Data

MongoDB · MySQL with SQLAlchemy · PostgreSQL · Firebase

Cloud & Integrations

AWS S3 · PayHere & Genie payments (sandbox → production)

Deployment

Vercel — incl. working around its lack of static IPs for payment gateways

Ways of Working

Git · Jira · Agile ceremonies

03The Work

Case
files

01

SmartLMS

versioned submissions & ai feedback

Final-year research project — SLIIT · Aug 2025 — Present

A microservices-based Learning Management System built to improve programming education through version-controlled submissions, AI-powered plagiarism detection, and real-time feedback.

I’m the sole developer of the AI-enhanced submission, feedback and version-control module within a three-person capstone team — my teammates built the project-management and analytics modules.

91%
True-positive rate

Ensemble plagiarism-detection engine — TF-IDF, Jaccard similarity and N-gram Dice coefficient — at an 8% false-positive rate, with real-time WebSocket risk alerts.

<5s
Feedback latency

Average response time for LLM-powered, real-time essay feedback.

23→6%
Contradictions

Feedback contradictions cut by a purpose-built consistency-enforcement engine.

−35%
Grading time

Instructor grading time reduced through AI-assisted grading.

SmartLMS answer editor with a live AI feedback panel scoring grammar, clarity, completeness and relevance, alongside similarity detection

Live AI feedback & similarity detection

01 / 05

Spring Boot · PostgreSQL · Redis · Next.js · TypeScript · LLM/NLP integration

github repo ↗
02

GreenRoute

intelligent waste management

Flutter mobile app — Android & iOS · Sep — Oct 2024

A Flutter mobile app for an intelligent waste-management system built to optimise garbage-collection routes, serving three roles from one codebase. Residents log garbage bags for collection, request special event pickups, report emergencies and track collection schedules; truck drivers get live collection routes drawn through the Google Maps Directions API with real-time location tracking, calendar schedules and QR-based fuel tracking; disposal officers record and review disposals.

Firebase handles authentication and real-time data sync across Firestore and the Realtime Database, while a Cloud Functions backend pushes FCM notifications — alerting drivers ahead of route starts — to improve efficiency and cut fuel consumption.

Flutter · Dart · Firebase Auth · Firestore · Realtime Database · Cloud Functions · FCM · Google Maps & Directions API · QR

github repo ↗
03

Villa Management System

villa operations management

Group project — 8-person team · Feb — May 2024

A web platform managing the day-to-day operations of a villa resort — room reservations and guest check-ins, service requests and customer support, staff scheduling and employee management, plus restaurant, events, transport and billing modules — deployed as a WAR on Apache Tomcat.

Second-highest contributor of eight: I built large parts of the core domain layer (the JPA entities, repositories and services behind reservations, rooms, staff, events and payments), the villa packages module, and much of the JSP frontend.

Java 21 · Spring Boot 3 · Spring Data JPA · MySQL · JSP · HTML/CSS/JS · Apache Tomcat

github repo ↗
04

simple-RAG

retrieval-augmented generation, built by hand

Solo practice build — full-stack RAG pipeline · Jul — Aug 2026

A full-stack Retrieval-Augmented Generation system built entirely from scratch — no LangChain, no LlamaIndex — to understand retrieval at a mechanical level: custom chunking with configurable overlap, a from-scratch NumPy vector store, and hybrid search combining BM25 and semantic embeddings via Reciprocal Rank Fusion (RRF).

A relevance threshold explicitly refuses to answer when nothing in the corpus is actually relevant, instead of confidently returning irrelevant chunks. Generation runs locally and for free via Ollama — no paid LLM API anywhere in the pipeline.

0.45
False-confidence peak

Highest similarity score a hand-written hashing embedder gave completely unrelated (nonsense) queries — versus ~0.07–0.11 for real semantic embeddings on the same queries, confirmed by direct comparison.

Python · FastAPI · sentence-transformers · rank-bm25 · NumPy · Ollama · Next.js · TanStack Query

github repo ↗
05

Weekly Report Generator

versioned team reporting with ai summaries

Solo full-stack build — Spring Boot & Next.js · Sep 2026

A weekly team-reporting platform that replaces status emails with an auditable workflow. Members file structured weekly reports; managers approve them or send them back with a comment, which forks a new editable version while every earlier submission and manager comment is kept as immutable history. Managers get a dashboard with compliance metrics and four charts, a cross-member comparison of blockers and achievements, and per-member profiles.

A Google Gemini assistant answers questions and writes a weekly summary, grounded only on a server-built digest of the reports that manager is allowed to see. Sign-up is approval-gated, JWTs can be revoked through a tokenVersion claim, login and password reset are rate-limited, and the reset flow never reveals whether an email exists.

281
Automated tests

139 backend tests (JUnit 5, MockMvc, Testcontainers on a real PostgreSQL) plus 142 frontend tests (Jest, React Testing Library).

34
REST endpoints

Across 9 controllers, serving three roles — member, manager and admin — each with its own purpose-built experience.

11
Database tables

Owned entirely by five Flyway migrations, with partial unique indexes enforcing one key blocker and one key highlight per version.

3
Access-control layers

A JWT filter, @PreAuthorize role gates and row-level checks inside services — a member can never read another member's reports.

Java 21 · Spring Boot 4 · Spring Security · PostgreSQL · Flyway · Testcontainers · Next.js 15 · React 19 · TypeScript · TanStack Query · Tailwind CSS · Recharts · Google Gemini

github repo ↗
04The Arsenal

What I
build with

Next.jsReactReduxTypeScriptTailwind CSSTanStack QueryLLM IntegrationRAGNLPGroqGeminiOllamaSpring BootJavaMongoDBPostgreSQLMySQLThree.jsDockerCI/CDGitAWS S3FlutterFirebaseOAuthTestingSEOAccessibility
05Get In Touch

Start a
conversation.