§— Portfolio

Built by our students.

Real production work shipped inside Kayalas TechLabs. Every case study below was architected, engineered and deployed by a student or cohort — not a demo, not a clone.

WorkPulse

Real-Time KPI & Time Management System

B2B SaaS · Internal Productivity Platform

ReactTypeScriptNode.jsExpressMongoDBSocket.io

The Problem

Managing task logs and team velocity in hybrid development teams often leads to stale data or administrative overhead. Standard tracking tools rely on periodic HTTP polling, which causes delayed updates or heavy read/write loads on the database. The objective was a real-time tracking interface capable of handling 50+ concurrent websocket sessions with update delivery under 50ms.

Architecture & System Design

Persistent WebSockets via Socket.io to eliminate HTTP handshake overhead for repetitive log events.
MongoDB schema with compound indexes on { userId, date } and nested session_logs sub-documents — keeping query retrieval under 15ms.
Optimistic UI updates: React state changes immediately; failed writes trigger a callback rollback with a user alert.

Engineering Challenges & Resolutions

Challenge

Server CPU usage spiked during testing because duplicate Socket client instances were created by React components on every state re-render.

Resolution

Centralised the socket connection inside a React Context Provider and added a clean-up hook to disconnect on unmount. Idle CPU dropped from 45% to under 5%.

Risk Guard

Loan Prediction & Risk Assessment System

Fintech SaaS · Machine Learning Classifier

Python (Flask)ReactChart.jsPandasScikit-LearnXGBoost

The Problem

Manual underwriting is slow, subjective, and error-prone. Financial institutions need high-accuracy classification systems that assess default risk instantly — but standard ML models behave as black boxes, providing predictions without explaining the underlying risk factors.

ML Pipeline & Technical Decisions

Imputed 15% missing data using median values to prevent outlier skewing; categorical variables encoded via Target Encoding to keep feature columns within limit.
Tested Random Forest and XGBoost; XGBoost selected at 92% validation accuracy and F1-score 0.90 — handles class imbalance more effectively.
Removed features with correlation > 0.85 (redundant loan duration calculations), cutting training time 15% with zero accuracy loss.

Engineering Challenges & Resolutions

Challenge

Loan officers rejected early models because they could not audit the rationale behind a high-risk score — making it unusable for regulatory compliance.

Resolution

Configured the Flask API to compute SHAP-based feature-importance contributions per prediction. The React frontend visualises these weights with Chart.js, showing underwriters exactly which factors drove the score.

Overseas CRM

Study & Work Abroad Management System

EdTech · Overseas Consultancy Management

ReactNode.jsExpressDatabase

The Problem

Managing student applications, visa documentation, and communication between agents and students is complex and error-prone. Traditional systems lack transparency and real-time tracking, leading to delays and poor user experience. The objective was a CRM platform tailored for overseas education and work consultancy workflows.

Architecture & System Design

Student Application Tracking — modules that follow each student from submission to visa approval.
Document Management — secure storage and retrieval of passports, academic records, and visa files.
Agent Dashboard — role-based views for monitoring assigned students, pending tasks, and statuses.
Scalable Backend APIs — RESTful endpoints engineered for large datasets and concurrent users.

Engineering Challenges & Resolutions

Challenge

Managing large volumes of documents while keeping access fast and the system performant under concurrent load.

Resolution

Optimised file handling with structured storage references and lazy loading techniques — preserving responsiveness as the document base scales.

Insurance CRM

Policy & Customer Management System

InsurTech · CRM for Insurance Agents

Full-StackReactNode.jsDatabase

The Problem

Insurance agents often juggle multiple clients, policies, and renewal schedules manually — leading to missed renewals and poor engagement. The goal was a CRM that simplifies policy tracking and strengthens client relationships.

Architecture & System Design

Policy Management Module — centralised system storing every policy with detailed metadata.
Customer Tracking — complete client profiles with policy history, communication logs, and status updates.
Automated Renewal Alerts — notification system reminding agents and clients of upcoming renewals.
Optimised Data Structuring — schema tuned for quick retrieval by expiration date and customer ID.

Engineering Challenges & Resolutions

Challenge

Ensuring timely and reliable delivery of renewal alerts without silent failures.

Resolution

Implemented scheduled cron-based background jobs with fallback retry mechanisms — guaranteeing alert delivery even when downstream services lag.

Ready to ship work that recruiters actually open?

Talk to Admissions↗