The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Back to feedPost
Wrote a bash script to analyze how often Hubstaff (the time-tracking software gig platforms use) takes screenshots of your screen. Turns out the "random" intervals aren't random at all, there's a consistent upper bound and average gap. Small project, but it's the same instinct I bring to Innerhouse Data: don't just take a platform's word for how it works, go verify it yourself.
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Related posts
That's what we call : Iteration 😃
Building a modular AI-powered SaaS platform is far more about architecture than writing code. 🚀
I've always believed the hard part of building SaaS isn't shipping screens—it's designing a system that can keep evolving without collapsing under its own complexity.
This dashboard started as a simple idea: consolidate business operations into a single workspace. It quickly became an architecture problem. ⚙️
The frontend was built around a modular component system where every feature behaves as an independent capability rather than another page bolted onto the application. That decision paid off later when adding new business domains without rewriting existing flows.
The backend was where the real engineering happened. 🏗️
Instead of tightly coupling business logic with APIs, I pushed the core into service boundaries with clear contracts, making each domain independently deployable and far easier to reason about. Every expensive operation runs asynchronously through event-driven workflows, keeping the request path lean while maintaining eventual consistency across the platform.
Real-time updates were another challenge. ⚡
Polling was never an option. The dashboard streams state changes over persistent connections, with optimistic UI updates, cache invalidation strategies, and fine-grained synchronization to avoid unnecessary renders. The goal wasn't just "real-time"—it was deterministic state across multiple clients without creating race conditions.
Then came the AI layer. 🤖
The biggest mistake I see is treating AI as another widget.
An LLM can generate endless observations, but most of them are operationally useless.
The difficult engineering problem isn't inference—it's context engineering.
The system retrieves only domain-relevant operational signals, enriches them with structured business context, applies retrieval pipelines, and constrains generation so the model produces recommendations with actual decision-making value instead of confident-sounding noise.
Every AI response is grounded in live operational data rather than generic prompts.
On the data side, the reporting layer was optimized to avoid analytical queries impacting transactional workloads. Materialized views, selective indexing, query planning, and aggressive caching keep KPI aggregation predictable even as datasets grow. 📊
The UI intentionally stays quiet. 🎯
No flashy animations. No unnecessary interactions.
Information density, visual hierarchy, and latency matter far more than decorative effects when people spend hours inside a product.
Looking back, the hardest part wasn't React, PostgreSQL, or integrating AI.
It was designing boundaries that let every layer—UI, services, data, and intelligence—evolve independently while still feeling like one cohesive product.
That's the kind of engineering work users never notice.
And that's usually a sign you got it right. 💡
What's the most challenging architectural decision you've had to make on a production system?
#SoftwareArchitecture #SystemDesign #FullStack #SaaS #AIEngineering #DistributedSystems #React #PostgreSQL #NodeJS #Python #ProductEngineering #WebDevelopment
Mimir — the AI execution layer for software
Build story
I kept hitting the same wall with AI agents: they can tell you what to do, but the moment you let one actually touch a real system — payments, orders, refunds — you're one hallucinated call away from a disaster. The demo is easy. The safety is the whole product. So I built Mimir around that problem instead of around the demo.
The core move is a single conversion: every API endpoint becomes one safe, callable operator. GET /orders becomes get_orders(). You point Mimir at an OpenAPI spec — or just paste the docs — and the model reads it, normalizes it, and commits it as a set of typed, validated tools an agent can run.
Three decisions shaped the build:
Boundaries live in the runtime, not the prompt. Permissions, approval gates, and destructive-write flags are enforced in code. You can't jailbreak a policy that was never a sentence. The agent is able to act, unable to overreach.
The model never sees the key. Credentials are vaulted and injected at execution time. The LLM operates the system without ever holding the secret to it.
Every action is signed and logged. Runs are stamped to an operator and written to an audit trail — so "the AI did it" is always answerable with who, what, and when.
The part I'm most proud of: I engineered it to run on a local 7B model, not a frontier API. That forced real discipline — small-model-first, code-enforces-not-prompts, eval-driven, structured output over brittle parsing, and a rule I made non-negotiable: the AI understands the docs and normalizes them; code only validates against known types. No hard-coding one company's naming conventions, because the next API won't share them. It's a deliberately-engineered system, not a wrapper around a chat box — and it's executed real refunds end-to-end.
Stack: Next.js 16 · React 19 · Tailwind v4 · TypeScript on the front; FastAPI · Postgres · LiteLLM → Ollama on the back. The interface is a custom design system I call "Operations Ledger" — carbon, bone, and one acid accent; Archivo Black; flat brutalist. It's meant to feel like an instrument panel, not a landing page.
Software you can operate by intent — safely.
#SoftwareArchitecture #SystemDesign #FullStack #SaaS #AIEngineering #DistributedSystems #React #PostgreSQL #NodeJS #Python #ProductEngineering #WebDevelopment
Trending
Claude
Claude has entered the design space. How are you using Claude Design?
Contra University
Learn from expert creatives how to earn more using next-gen AI tools.
creativeaiflow
Creative AI workflows are evolving. What tools do you use, and what are their strengths and weaknesses?
freelancerlife
Freelancer life is wins, pivots, and everything in between. What’s yours right now?