From 'Human Google' to Autonomous Agent: Why I Built Aimpler KnowAI
How a decade-old habit of sharing industry insights evolved into an automated, token-efficient AI intelligence hub.
About a decade ago, I was leading sales and marketing for a VC-funded, early-stage hotel tech startup. I developed a reputation that, surprisingly, became my secret weapon for easily landing meetings with key stakeholders.
I had a relentless habit of reading every top e-magazine covering hospitality tech. It wasn't just for me—I loved sharing these links with colleagues and friends. More importantly, I started dropping little "insight snippets" into meetings and calls with clients and stakeholders. They absolutely loved it. We'd exchange ideas, debate trends, and it shifted the dynamic from a standard sales pitch to a high-value strategic conversation.
My team eventually started making fun of me, dubbing me the office "Google" (haha!).
Some habits just don't go away. Fast forward to today, and I do the exact same thing with Artificial Intelligence. I am constantly reading, sharing, and discussing the latest breakthroughs. But there’s a massive difference now: the sheer pace of evolution in AI.
Trying to manually keep up with LLM benchmarks, GPU supply chains, agentic workflows, and global policies is like trying to drink from a firehose. It became nearly impossible to maintain my old habit manually.
So, I decided to automate myself. I built my oldest habit into an autonomous agent for everyone to use.
Enter Aimpler KnowAI.
The Mandate: 9 Pillars of Intelligence
I didn't want to build just another generic news aggregator. KnowAI needed to operate on a strict 12-hour update cycle, run completely autonomously, and provide high-signal intelligence across the specific categories that matter most to SaaS founders and operators:
The goal was to synthesize the noise into an "at-a-glance" dashboard that respects your time, just like those insight snippets I used to share in meetings.
The Engineering Challenge: Solving the "Token Tax"
The simplest way to build an AI news aggregator is to throw thousands of raw articles at a massive LLM like GPT-4o and ask it to summarize everything. That is also the fastest way to bankrupt a project with API costs.
To build KnowAI sustainably, we had to architect a tiered, "Zero-Token Discovery" strategy.
Tier 1: The Zero-Token Filter (Python)
Before an AI model ever sees a headline, our backend Python engine performs ruthless, cost-free signal analysis on hundreds of incoming RSS feeds. By using weighted keyword scoring (e.g., prioritizing "Launch" and penalizing "Opinion") and enforcing source diversity, the script filters out roughly 90% of the noise before it ever touches an LLM.
Tier 2: Precision Batching (The Agent Layer)
Only the "Top 5" highest-signal items survive Tier 1. These are sent in batched prompts to a highly efficient small language model (GPT-4o-Mini). The agent doesn't just summarize; it is instructed to structure the data into strict JSON, assign numerical "Impact Scores," and tag content for relevance.
This tiered approach allows KnowAI to achieve near-identical quality to expensive models at a 95% reduction in token costs.
The Result: "At-a-Glance" Intelligence
The output of this autonomous pipeline is a dynamic Bento Grid interface designed for rapid consumption. The frontend architecture uses specialized layout logic to categorize news visually. You might see a wide, four-tile spread covering complex EU AI Act developments sitting just above a focused square highlighting a breakthrough in NVIDIA’s GPU supply chain.
The entire system runs on fully automated CI/CD pipelines via GitHub Actions. It rebuilds the dashboard with fresh intelligence every morning and evening, completely hands-off. Public users simply read from a static JSON file, meaning zero API overheads and lightning-fast load times.
I built KnowAI to solve my own information bottleneck and keep my favorite habit alive. It’s a testament to the power of modern agentic workflows—moving beyond simple chatbots to systems that plan, execute, filter, and present complex data autonomously.
Level Up: The Native App Experience
We didn't stop at just a responsive website. Aimpler is now a fully installable Progressive Web App (PWA).
By leveraging modern browser capabilities like Service Workers and web manifests, you can now "install" Aimpler directly to your home screen or dock without an app store visit.
- Zero-Friction Install: A smart banner detects your device and offers a one-tap install (Android/Desktop) or guides you to "Add to Home Screen" (iOS).
- Offline Intelligence: Going into a subway or flight mode? No problem. The app caches the latest intelligence feed, so your dashboard works instantly, even without an internet connection.
- Native Feel: Launching from the home screen removes browser chrome, giving you a full-screen, immersive experience.