Holgrex
Senior Software Engineer — shipped 15+ production applications across healthcare, real-time analytics, AI SaaS, and mobile, serving 500K+ users
Role
Senior Software Engineer — Architecture, Design System, AI integration
Tech Stack
React, Next.js, TypeScript, Tailwind CSS, Storybook
Tools
Figma, GitHub, Vercel, MongoDB, Supabase
Focus
Healthcare, Analytics, AI SaaS, Mobile
Status
Ongoing
Intro
At Holgrex, I lead frontend architecture across a technology consultancy that ships production software for clients while building its own products in parallel. The work spans mobile, web, and AI domains across every client engagement and internal product.
The work spans healthcare platforms, operational analytics dashboards, an AI-powered SaaS product with real-time data streams and tool calling, and a shared design system that now underpins every project. Across 15+ shipped applications, the recurring challenge is the same: deliver fast, maintain quality under deadline pressure, and leave a codebase the next engineer can reason about.
Design System & Component Library
Early in my time at Holgrex, each project was building the same primitives independently — buttons, form inputs, modals, data tables — with subtle inconsistencies that accumulated into visual debt. I built a shared component library using Storybook and Tailwind CSS that became the single source of truth across all projects.
Storybook-first
Every component is developed in isolation with documented variants, accessibility annotations, and interaction tests. QA happens at the component level before any page integration — catching regressions without running a full application.
Tailwind tokens
Design tokens — color, spacing, typography, shadow — are defined once in Tailwind config and consumed everywhere. Theming a new client brand is a config change, not a grep-and-replace across hundreds of style declarations.
Hours, not days
New projects bootstrap in hours rather than days. Engineers spend time on product-specific logic, not re-implementing a date picker for the fourth time.
Healthcare Platform
One of the most demanding engagements was a healthcare platform where clinicians needed fast access to patient data across multiple care modules. The existing system had accumulated years of unoptimized API calls — pages took 8–12 seconds to reach an interactive state, which in a clinical setting is unacceptable.
The fix was not a simple cache layer. The data graph had deep interdependencies: a patient record pulled demographics, active medications, recent lab results, upcoming appointments, and insurance status in five sequential round trips. I restructured the fetching layer to parallelize independent requests, colocate data fetching with the component tree using React Query, and preload predictable navigation targets on hover.
The result was a smooth, responsive interface that loaded critical patient data in under two seconds. Clinicians reported meaningfully less friction in their day-to-day workflows — a measurable outcome in a domain where time directly affects care quality.
Real-Time Analytics Dashboards
Several Holgrex clients needed operational dashboards that surface live metrics — sales pipelines, logistics throughput, support queue depth — with data refreshing in real time. The challenge was keeping the UI fast and readable under a continuous stream of updates without triggering unnecessary re-renders.
WebSocket + optimistic state
- Persistent WebSocket connection streams delta updates rather than full dataset refreshes
- Local state updates optimistically on user action; server confirms or rolls back
- Stale-while-revalidate pattern keeps charts populated during connection interruptions
- Virtualized rows in large data tables — 10,000-row grids render without jank
Chart & filter architecture
- Chart components are decoupled from data-fetching — each receives typed props and renders independently
- Global filter context synchronizes date range, segment, and granularity across all panels
- Derived metrics computed client-side to avoid round trips on every filter change
- Export pipeline serializes visible data to CSV/XLSX without a server round trip
AI-Powered SaaS Platform
I led frontend development on an internal AI SaaS platform that integrates real-time data streams with LLM tool calling — allowing users to query live operational data in natural language and receive structured, actionable responses. This was the most technically complex engagement at Holgrex.
Streaming UI
AI responses stream token-by-token. The UI renders partial text progressively, parsing structured tool-call outputs mid-stream and rendering chart or table components before the full response arrives — eliminating the blank-screen wait.
Tool calling
The LLM calls typed tools — query-data, filter-results, generate-chart — that execute server-side and return structured payloads. The frontend treats tool results as typed events, not raw text, making the rendering layer deterministic and testable.
Live data context
Real-time metrics are injected into the LLM context window at request time — the model always reasons over current data, not a static snapshot. Context size is managed by summarizing historical turns rather than truncating them.
Mobile & Cross-Platform
Several Holgrex clients needed their web products mirrored on mobile without the cost of maintaining two separate codebases. I built cross-platform mobile applications using React Native, sharing business logic and data-fetching code directly with the web layer — only the rendering primitives differed.
This approach — a shared TypeScript core consumed by platform-specific UI shells — meant bug fixes and API layer changes propagated to both platforms in a single commit. Performance on lower-end Android devices was a specific focus: gesture interactions were moved off the JS thread using Reanimated, keeping animations at 60fps regardless of JS workload.
Industries
Across 15+ shipped applications, Holgrex work has touched verticals with meaningfully different technical and compliance requirements.
Healthcare
Patient data access, clinical workflow tooling, multi-module platforms. Performance is not optional — slow tools cost clinician time and affect care quality.
Logistics & Operations
Real-time dashboards tracking shipment status, warehouse throughput, and SLA breach risk. High update frequency with virtualized grids handling 10K+ rows.
SaaS & AI Products
AI-powered platforms with LLM tool calling, streaming responses, and plan-based access control. Architecture must handle both rapid iteration and production reliability.
Education
Course platforms and school management systems requiring accessible, mobile-first interfaces for student and administrator workflows at scale.
Engineering at Scale
Running concurrent client engagements means that architectural decisions compound. Shortcuts taken on project three show up as bugs on project seven. The design system, shared TypeScript utilities, and consistent patterns around data fetching and error handling are not abstractions for their own sake — they are the operational answer to shipping 15+ applications without a proportional increase in headcount.
Every deployment pipeline runs through CI with type checking, lint, and automated tests before merge. Monitoring with structured error capture means production issues surface in under five minutes. 99.9% uptime across the portfolio is a product of repeatable process, not heroics.
The most durable contribution is not any single application — it is the standard that makes the next application faster, safer, and more maintainable than the one before it.
That's Holgrex — client-driven engineering at scale.
← Back to Portfolio