Everyone else is doing AI. That’s why we just put up a beta of our new Intellyx Brain Chat in order to stay ahead of the curve in the competitive technology analyst market. Yes, you can try out the first iteration of it now at https://chat.intellyx.com.
Here at Intellyx, we cover digital transformation broadly, which means we buck the trend of categorizing all of the vendors into buckets. Over the last 11 years, we’ve written more than 5,000 articles that connect the dots between multiple technology solutions, including many BrainBlogs, whitepapers, briefs, news placements and Cortex columns just like this one.
A few years ago, we launched the Thought Leadership Finder to surface the value of all of our articles in a browseable form. Yes, it’s just an index of our entire content database, with each piece tagged with one or more relevant solutions.
Interestingly, we didn’t need to retire the categories we carried on from the early 2010s like SD-WAN and DPA. Even if they are still useful, they just fade away, as the newer cloud native development and AI-based solutions gradually got more attention. End customers changed their expectations, and solution vendors evolved their positioning with each briefing.
Why we started down this path
This July, me and my brilliant wife Elena and our two girls were in Europe for a whole month. The first two weeks in Poland, I still worked in Poznan and took calls, visiting her family and then getting out to the mountains in the south near Slovakia, then mostly ejecting from work and visiting all of the Scandinavian countries we had never been to—Denmark, Sweden, and even remote northern Norway, where it was warm enough (51F) to swim in the Arctic Ocean.
One cool aspect of having a full-stack developer as a companion, we had plenty of time for conversations as we were moving around and the kids were on their devices to talk about the future. How will developers and executives, and even analysts stay relevant in today’s AI-driven economy? The capabilities advance so fast, everyone might think information-related work will become obsolete.
On the other hand, AI is generating new forms of technical debt at a rate we can’t even comprehend, and we’ll need people to untangle it for business use. I did a pretty deep dive on observability for AI-generated development in SiliconANGLE a few months back, and talked to the newly-minted “Head of AI” for several strong vendors. Each had great insights, as well as totally unique methods for getting predictable, deterministic results out of a non-deterministic system.
Fortunately, we’re not building an enterprise-grade system, we just wanted to have a version of our analyst content that readers can talk to. And lucky for me, Elena was game to try building one!
Avatar: the first plan
I’ve seen a few convincing conversational AIs with human-like avatars that you can actually talk to on a video chat screen, and they are getting pretty quick, down to sub-second responses in many cases. Some can perceive voice inflection and facial cues to respond to the human user’s approval or disapproval. Many of the avatar models in use today are attractive-looking females for some reason, but you can even customize them to look exactly like yourself.
We started to tinker with demos, but found some different issues with representing ourselves through an avatar. One solution had an up-front setup services project of 2-4 weeks. Another limited the amount of knowledge base size to around 100 pages worth of writing, which would probably work OK for a CSR call script, but not for our massive content. The next one charged per video minute, so it definitely wouldn’t fit into our low-overhead analyst business model to have a ballooning SaaS bill if it caught on.
None of these constraints would stop a regular company from making an investment in virtual agents, if it could deliver ROI versus doing things the old fashioned way. In the end, it’s just a demo app for us anyway, so we decided to stick to a regular text-based chat format and try to use open source or low-cost tools, models and hosting, to avoid opex.
Loading up our RAG, with relevancy and tools
The golden rule of machine learning and AI has always been that your model is only as good or as useful as the data you train it with. As an analyst firm, we own a lot of 100% human-generated expert content, so that represents a pretty good corpus of information to start from.
To start, I loaded a bunch of my whitepaper PDFs and BrainBlog docs into a Google Drive folder, and Elena began running imports into a Postgres database (hosted on Supabase) via a workflow she set up in n8n. The first “JE content only” versions of the chat were, as expected, pretty hit and miss. Often, the LLM would go out of range and pull general information from the internet, unless specifically prompted not to.
Elena used some AI in her web scraper and encountered similar “hallucinations” when asking the AI to find full article links for content posted on external sites—the AI occasionally made up random URLs. This highlighted how, unless explicitly instructed otherwise, LLMs almost always generate fabricated answers.
Once the chat responses started to stabilize, and we found it was answering when it should, and not answering when it shouldn’t, Eric and Jason uploaded all their own content to the Drive folders, where it automatically gets hoovered up by an automated process. Additional content was pulled in from our externally hosted columns and articles at The New Stack, SiliconANGLE, DZone, SD TImes and so on.
Elena began the RAG backend work with n8n, a no-code tool, and used Python scripts for crawling and scraping content both on our website and articles hosted elsewhere. However, she soon shifted to a custom-coded solution built mainly with Node.js and Supabase/Postgres for the RAG retrieval backend, finding it surprisingly straightforward to develop a RAG backend from scratch. She still maintains an n8n workflow to process new white papers from Google Drive folders, alongside a Python cron job that leverages Crawl4AI to crawl and scrape the website for new content.
For more in-depth detail of this and other development challenges and workarounds, check out Elena’s nice recap of the Intellyx Chat development story here on Medium.
What about your ‘No AI Guarantee’?
Here at Intellyx, we never use AI to generate any part of our content. We don’t even use AI note takers to summarize meetings, because that would color our impressions. Even more than standard SEO, GenAI search engines will pass over AI-generated content in favor of original content by known authors.
Besides, I like taking my own notes anyway, it’s how I retain information.
Thus, creating a fun chat interface based on our own human-generated content isn’t breaking this pledge, it’s reinforcing it. Also cool, if I want to find out the last three times we mentioned a topic, Brain Chat returns attributions and dates under each response. Why search documents and websites, which are usually poorly tagged?
The Intellyx Take
Where do we go from here? I hope you’ll try Intellyx Brain Chat out, and send us your suggestions, complaints, or queries about using the words you get out of it at pr@intellyx.com. It is definitely not a replacement for talking to us, we are much better in person.
Going forward, there’s an automated routine to continue updating the RAG with new content as it is added to our site. Some of my early beta testers wanted each of us to answer some questions about ourselves personally. Why not find out about my favorite beers?
Maybe we’ll even revisit the human-looking video avatar idea someday soon, but rather than using it for incoming briefings, we’ll send our virtual selves to attend group webinars which we don’t have any spare time for.
You can’t begin before you start, so until tomorrow, now is the future.
Copyright ©2025 Intellyx B.V. Intellyx is an industry analysis and advisory firm focused on enterprise digital transformation. Covering every angle of enterprise IT from mainframes to artificial intelligence, our broad focus across technologies allows business executives and IT professionals to connect the dots among disruptive trends. As of the time of writing, none of the organizations mentioned in this article is an Intellyx customer. No AI chatbot was used to write this article. Image source: Jason English (photo from Sommarøy, Norway).
