Selected work

The Challenge

Employees had constant questions about leave policy, benefits, and internal procedures — but the answers lived buried inside long HR policy PDFs and intranet pages nobody wanted to dig through. In practice, that meant a steady stream of repetitive questions landing on HR's desk instead of employees just finding the answer themselves.

Our Approach

A generic chatbot wasn't good enough — answers about leave entitlement or reimbursement policy need to be grounded in each company's actual documents, not a plausible-sounding guess. We designed a retrieval-augmented generation (RAG) system so every answer is retrieved from, and traceable to, the real policy text.

What We Built

An internal RAG assistant built on Kotlin and Micronaut, using Meta's Llama-3.3-70B-Instruct model to generate answers grounded in each company's own HR policy documents. Employees ask questions in plain language and get an instant, accurate answer instead of searching a PDF or waiting on HR. The same architecture extends naturally to other departments or a company-wide knowledge base.

Result

The Challenge

Orders arrived by phone, email, and fax, then were re-typed by hand into three separate systems: an order log spreadsheet, QuickBooks for invoicing, and a separate inventory tool for the warehouse. Double-entry errors were routine, and staff spent hours each day just reconciling numbers across systems — with no one able to say, at a glance, what stage any given order was actually at.

Our Approach

Before designing anything, we spent two weeks shadowing the order desk and mapping the full order lifecycle end to end. The real bottleneck wasn't any one tool — it was that none of them talked to each other, so every handoff between people was also a handoff between disconnected systems.

What We Built

A centralized order management web app that became the single entry point for new orders, integrated directly with QuickBooks for invoicing and the existing warehouse inventory tool for live stock levels, with a shipping carrier integration handling label generation and tracking updates automatically.

Result

The Challenge

All four clinic locations ran on a decade-old desktop scheduling application: no online booking, no visibility across locations, and no automated reminders. No-show rates were high, and front-desk staff spent most of the day on the phone just confirming and rebooking appointments.

Our Approach

We focused the design phase on the front-desk staff's actual daily workflow rather than simply moving the old software into a browser. That surfaced exactly where confirmations and reminders were falling through the cracks between locations.

What We Built

A web-based scheduling platform with real-time availability shared across all four locations, integrated with the clinic's existing EHR system for patient records and an SMS reminder service that sends automated appointment confirmations without staff involvement.

Result

The Challenge

The retailer ran four disconnected tools — inventory management, an online storefront, a third-party fulfillment provider, and a customer support helpdesk — with staff manually re-keying data between them. Nobody had real-time visibility into stock or order status, and reconciling the four systems ate up a full day of staff time every week.

Our Approach

Each of the four tools worked fine on its own, so replacing them wasn't the answer. We designed a middleware integration layer instead — one that would keep all four systems in sync automatically, without disrupting the workflows staff already knew.

What We Built

A real-time API integration layer syncing inventory, orders, and fulfillment status across all four platforms, plus a lightweight analytics dashboard giving leadership a single, always-current view of sales and stock health.

Result