Our work

Live demo · HR system with AI assistant (POC)

Inori HR + AI HR assistant

We built an HR system for Malaysian SMEs with an AI assistant inside it: it answers from the employee handbook and the signed-in user's own records, and it prepares leave and claims for one-click confirmation.

• Employee records: searchable profiles, departments, a company directory, and probation, contract and leaver tracking.
• Leave on Employment Act entitlements by length of service, with separate medical leave and state public holidays never counted.
• Attendance with clock-in and clock-out in Malaysia time, late detection and monthly reports per person and per team.
• Expense claims with manager approval, a mileage helper and reimbursement on the next payroll run.
• Payroll with EPF, SOCSO, EIS and PCB, pro-rated joiners and leavers, printable payslips and a statutory summary.
• An AI assistant that acts with the signed-in user's permissions, asks for confirmation before any change and logs every action.

• Next.js
• TypeScript
• PostgreSQL
• Prisma
• OpenAI models
• tool calling

Try asking the assistant

• Employee login“How much annual leave do I have left?”
• Employee login“Which public holidays are coming up?”
• Employee login“What does the handbook say about medical leave?”
• Manager login“Approve the oldest pending leave request”
• HR admin login“Summarise the last payroll run”

Demo logins: hr.admin@inori-demo.my (HR admin), manager@inori-demo.my (manager), employee@inori-demo.my (employee). Password Demo@1234 for all. Demo data resets every hour. Please don't enter real personal information.

Scope: fictitious company and staff; statutory figures are demonstration-grade, not payroll advice. Can be installed on your own server, with the assistant on a cloud model or a private model on NVIDIA hardware.

Live demo · Document Q&A (RAG) POC

Ask your documents

Upload a policy, a manual or a price list and ask questions in plain language. Answers come only from your documents, with the passage each answer was taken from.

• Takes PDF, Word (.docx), TXT, Markdown and CSV files: up to 5 MB each, five files per workspace.
• Hybrid search matches every question by meaning (vectors) and by keywords (full text), then merges both rankings.
• Every answer cites the exact passage it used, with the file name and page, and the Sources panel shows the text.
• When the answer isn't in your files, the assistant says so instead of making one up.
• Each visitor gets a private workspace that is deleted automatically after 30 minutes; original files are never stored, only their extracted text, and that goes with the workspace.
• Runs fully on-premise when needed, with a private model on an NVIDIA RTX workstation or DGX Spark, so documents never leave the office.

• Python
• FastAPI
• PostgreSQL
• pgvector
• OpenAI models

Try asking on the demo

• Price list (CSV)“How much does a website AI chatbot cost, including the monthly fee?”
• Word document“What is the response time for a P2 IT ticket?”
• Support SOP (Markdown)“Can a customer get their deposit back if they cancel a website project?”
• Asked in Malay“Berapa hari cuti sakit untuk staf yang bekerja 4 tahun?”
• Deliberately not in the documents“What is the company's policy on the CEO's bonus?”

The Malay question is answered in Malay from English documents. The CEO-bonus question has no answer in the sample files, so the assistant should reply that the answer isn't in the documents rather than invent one.

Sample files for a fictitious company are on the demo page. Please don't upload confidential or personal data.

Scope: a public proof of concept. The sample documents describe a fictitious company, and an AI answer can still be wrong, so check the cited passage before relying on it.

Proof of concept · Public demo

Carton ordering portal

We built a carton-ordering proof of concept for a packaging manufacturer, shown under a fictitious brand: a public product site, a customer portal and an admin console in one application, plus a public Build your box designer that prices a custom carton as the visitor designs it.

• Public Build your box designer: 7 box styles with product photos, photo, dimensions and flat-blank previews, and an instant estimate with price breaks that show how a larger run lowers the unit price.
• Estimates come from a versioned price book staff can edit and roll back, with a best-buy quantity suggestion; bookings keep the exact design and estimate.
• A public product catalogue leads into a specification-based carton booking flow.
• Customers submit specifications, review and accept quotations, and follow order status.
• Staff quote bookings, move orders from submission through delivery, and print job sheets.
• Inventory records receipts, issues and adjustments in a stock-movement ledger, with low-stock indicators and CSV export.
• Role-based customer/admin access keeps the right screens and actions with the right users, and the header changes to My portal or Admin console once signed in.

• Next.js
• React
• TypeScript
• Tailwind CSS
• PostgreSQL
• Prisma

Demo logins: admin admin@contoh.example and customer purchase@contoh.example. Password Demo@1234 for both. Demo data resets every hour. Please don't enter real personal or business information.

Scope: estimates use placeholder prices benchmarked on Malaysian market ranges and exclude SST, and the box-style photos are placeholder product images. This is a booking and quotation POC; no online payment is taken. The brand, company, address, people and orders are fictitious demo data. Branding, catalogue content and business data are demo material, not client performance evidence.

Concept build · Local-business landing page

Eco Fast Car Wash

We built a friendly, mobile-ready concept landing page that brings a local car wash's services, published prices, hours and directions into one place.

• Clear service cards and a vehicle-size price layout, with the published board's date and a reminder to check current prices at the counter.
• Weekly hours and a small open/closed indicator based on Malaysia time.
• Direct directions and business-contact links without an embedded map.
• An original visual identity and locally served, credited illustrative photography.
• Responsive, accessible HTML with keyboard focus styles and reduced-motion support.

• HTML
• CSS
• Vanilla JavaScript
• Static site
• JSON-LD

Scope: this is a concept demonstration, not a claim of a commissioned engagement or an owner-endorsed official website. The photos are illustrative stock, not on-site evidence.

Brand concept · OAuth-ready information site

Snuggle Paws

We built a warm dog-boarding brand concept with a home page, privacy policy and terms for Google OAuth-ready branding on a personal AI-assistant project.

• An original dog-focused identity, responsive home page and locally served photography.
• Service cards, a sample daily routine, FAQs and clearly labelled example pricing.
• Dedicated privacy and terms pages linked consistently throughout the site.
• Plain-language disclosures for intended Google Calendar, Gmail and Drive uses, including access revocation and Limited Use commitments.
• A static public site without sign-in, cookies, trackers or third-party scripts.

• HTML
• CSS
• Vanilla JavaScript
• Static site

Scope: Snuggle Paws is fictional demo branding, not an operating boarding client. This public site does not implement Google sign-in or an AI assistant. OAuth-ready refers to the branding and policy surface, not Google approval, certification or a completed OAuth flow.