.png)
2 hours ago
2
Vision
The Internet is Search.
Search engines have defined the way we travel accross the web for decades.
Some say that with AI all search engines will be replaced by Agents or that AI has completely solved the problem of search.
But let's be honest: That's entire crock of shit!
The Problem
The number of ways we can search the web today is infinite.
We can:
- Google it
- Ask ChatGPT
- Ask Perplexity
And many more considering how many search engines and AI search tools there are.
The problem is not the search
The problem is in the answer
The search is easy to automate due to its technical nature.
The process that every single search engine or search tools uses is literally the same (web scraping, text or link extraction and embedding, etc.).
But unfortunately, if you have used the tools, you'll know that even the answer is always the same.
It's either the links that Google gives you, or the text and citations that ChatGPT and Perplexity give you.
But all of those answers are relevant to the prompt/search you generated, they are just meant to give you an answer based on the aggregation of words and symbols you put together to trigger the search.
But what about YOU?
What about the goals YOU have that lead you to ask the query?
What about the flame that sprinkled YOUR curiosity?
Do you want more answers, or do you want more RELEVANT answers?
Approach
My view of search is the one of a personalized experience.
I don't just want the truth-grounded answer, I want that the answer reflects the question I asked.
I often use Perplexity to create roadmaps or study plans every time I want to learn something new.
The problem I always face is even if I craft the more specific prompt ever written, with role, context, guidelines etc., the answer almost seems to follow the spirit of "Let's give him the most amount of information we can find around the query", not "Let's find the most RELEVANT amount of information present in the web related to his question"
And this is, in my opinion why we still haven't cracked the search problem.
LLMs can hallucinate.
Search engines are too broad.
Perplexity gives you a lot of things you don't even need.
What's missing?
The personalization of search.
The empathy of search.
Asking a query and be able to say: "It just got me."
Scope
In the Github repository linked above you will find a little and unfinished draft of a Rails 8 AI-powered search assistant that answers questions using only information scraped from the web.
It runs a multi-stage pipeline: web search → scraping → vectorization → truth-grounded AI synthesis with citations. Responses must never hallucinate and must always include proper citations..
Core Principles
- Personalized Search: The search will be shaped around the goal and instruction provided to the assistant in order to receive the most personalized output possible
- Truth only: Synthesize from scraped content; never inject model “knowledge.”
- Citations everywhere: Inline citations like [1], [2] must appear next to claims.
- Robustness: Timeouts, retries, rate-limiting, fallbacks for scraping failures.
- Safety: Sanitize inputs and scraped HTML; validate URLs; store only non-sensitive data.
Conclusion
Thank You for having read this manifesto.
If you share the same vision ans values please consider leaving a star on the Githbub repository above and if you want to contact me, below you will find my X profile and Linkedln profile.
Thanks for your time!
Alessandro Aimar
