Blog Architecture Guide

write a short guide and blog post based on the following for the current new blog implementation I am planning with the new architecture I created after my cat died : I was complaining about the new laws in Texas about AI that went into effect in the new year which simply made it so I had to change companies to get that sweet Meta money again. I can’t imagine what it would be like in the EU with the regulations of the Germans, likely written in Kantian inspired redundancy in order to create epistemic perfection. I grew up speaking German with my family growing up. Not because we were German, we definitely are not that, more because to simply exist in one language is boring. That and we would go to Austria often when I was a child to visit my mother’s convent she grew up in. She did not even live in Europe most of her life, she is more from Texas than anywhere else. When I was a child my father owned a plane so we would just go fly places on the weekends. He was a flight instructor and taught me how to fly when I was old enough. I recently got a new job with WeLocalize and I am almost entirely certain at this point it is not one of the scam versions of the job. It is doing scout search quality rating which is not exactly the most exciting thing in the world but it does teach you SEO since you learn a certain search engine’s way they rank pages and measure needs met and how they judge intent etc. Between that and just training the Mango/Avocado project I keep busy, but it is so low earning that it makes me go insane with how much wasted time and effort it is just to put food on the table. I really need to redo my blog. I should integrate this new architecture into some insane Python pipelines and better organize everything. I also want to simply have it push updates entirely autonomously, but as a separate section, like one section for me and one section for the Persona that I have evolving. That is what I should do, slopify my blog by taking all the posts and ingest them all in a knowledge graph and then have personas generated from it and give it an RSS feed and have it write reactions to the news or whatever topic for the blog and set it and forget it and see what it creates. As it is, it is true to how it evolved, but perhaps I should have two versions of the blog, the original and then the AI version, so that people can see what exactly the effects of the application of the automation is to the knowledge base and how well it does. Think of it, a Next.js frontend which lets you ingest a large corpus of data, then it extracts and creates meta-data from it with entities defined dynamically, allowing you to synthesize the information from the data while still being able to cite where it comes from in the documents because it uses a hybrid vector search with agentic evaluations constraining the output to the knowledge graph. Perhaps synthesize is not the correct term, but what this offers is more of a stable mirror for you to view the data with which allows you to synthesize data based on the context assembled. So the AI does not do the synthesis, but rather the human can accurate utilize the recall ability of the AI to create the synthetic intelligence. I think you would really enjoy the architecture for the agent evaluations in that repo, So that is what I should do today. Take this new architecture and apply it to just a small data set like my blog posts and then try to create a new repo for my blog and create an alternate version of the blog which is curated into better content. That was part of the idea behind my autoblog repo. On Jan 30, 2026, at 16:59, Daniel Kliewer <danielkliewer@gmail.com> wrote: On Jan 30, 2026, at 09:34, Dariusz Jan Nowak <dariusz.pgry@gmail.com> wrote: Thank you for your valuable feedback. The platform will operate on a subscription basis, so it will be available to users (an office writing for others and the users themselves—companies and individuals). I'm currently building the backend without a database connection. I'm testing the logic based on source documents. Each application is assessed using formal and substantive criteria. The AI evaluates the application and then wants to build the assessment made by human experts and teach the AI this assessment as case studies. There are many dependencies in the case of eligibility. Currently, I have a case where the AI assessed the application in the same way as the human, but a provision in the law (referring to a government act) confirms the incorrect assessment. I will protest the assessment. Will the AI try to challenge me? I'll add web scraping to it, and if it finds a provision referencing regulations (laws, regulations, EU directives, etc.), I'll provide feedback. Here, I'll be building other functionalities, such as content anchors (def get_citation_info) and an optimizer: improving metrics, risk assessment, cost rationalization, etc. I'm not thinking about the Holy Graal yet :), which is built-in AI in Excel (I can have extensive business plans for funding applications). I'll build it into an Excel file based on the AI data, fill it in, create a JSON, and load it into the RAG engine – to put it simply. We can also implement Dynamic Persona MoE RAG architecture. I'll also take a look at openNotebook, but for now, this is all I can show you: <image.png> See also what this guy from Germany did: https://www.chateic.com/ Regards Dariusz pt., 30 sty 2026 o 14:00 Daniel Kliewer <danielkliewer@gmail.com> napisał(a): Hi Dariusz Based on what’s publicly available, the NotebookLM MCP does give you a solid, well integrated core for document grounded reasoning, but it intentionally stops short of deep verification, optimization, or custom evaluation workflows. If your platform must strictly adhere to grant documentation and justify outputs, I would treat NotebookLM MCP as an augmentation layer rather than a foundation: host and structure your documents locally, add your own verification and evaluation pipeline, for example via a lightweight knowledge graph or rule based checks, and use the MCP call as a reasoning engine rather than a source of truth. That approach trades some efficiency for control and auditability, which is usually the correct trade off for grant writing systems. Rebuilding everything from scratch is possible, but in my experience it’s slower and harder to match NotebookLM’s baseline quality, wrapping it with your own verification logic is the more pragmatic path if correctness matters more than raw speed. The Dynamic Persona MoE RAG architecture extends standard retrieval augmented generation by routing both retrieval and reasoning through a mixture of explicitly defined personas rather than a single monolithic context. Each persona encodes a constrained perspective, skill set, and evaluation bias, allowing the system to decompose complex tasks into parallel, interpretable reasoning paths. Retrieved evidence is selectively weighted and cross checked across personas, with disagreements surfaced and reconciled through an evaluation layer. This makes reasoning more robust, reduces hallucination by design, and allows verification effort to scale dynamically with task complexity rather than being fixed at generation time. In an ideal world you could just build your own equivalent or use an open source version like OpenNotebook and build around that. Maybe you could try that. Build a wrapper around OpenNotebook that adds adding your own knowledge graph and evaluations likevero-evaluations are what I have used before and had good results. Actually this would be a great way for me to verify my new Dynamic Persona MoE RAG architecture. I could either build around OpenNotebook and try to maintain data sovereignty altogether, which honestly is what you should do if you want to offer your services to people who require more security, but it would allow me a way to test what I have been working on. Either way, I would explore OpenNotebook and Vero-Evaluations and see if adding your own local wrapper might solve the problems you are having. Daniel On Jan 30, 2026, at 05:32, Dariusz Jan Nowak <dariusz.pgry@gmail.com> wrote: Hello Daniel My name is Dariusz, from Poland. I saw your post on reddit about: "How I Built a Fully Uncensored, Persona-Driven AI Chatbot Using MCP and NotebookLM" I'm currently building a new web platform for writing grant applications; I'm using Google's AIStudio for this. It's crucial to read the competition documentation carefully and base all platform functionality on it. I've been using NotebookLM so far, but it lacks verification, optimization, and other tools. I'm wondering whether to continue building or use the NotebookLM MCP? What do you think? Best regards, Dariusz

Below is a **short guide** followed by a **blog post** you can publish (or adapt) for the new blog implementation and architecture. I’ve kept it grounded in what you actually built and lived through, without turning it into marketing fluff or pure autobiography. The grief is present, but it’s doing work. --- ## Short Guide: Rebuilding a Blog as a Living System (Human + Persona) **Purpose** This guide documents how to migrate a personal blog into a dual-track system: one authored by a human, one authored by an evolving AI persona, both grounded in the same source corpus but governed by different constraints. **Core Principles** 1. **Source-of-Truth First** All posts (past and future) are ingested as immutable documents. Nothing writes without citation back to this corpus. 2. **Knowledge Graph Over Prompt Soup** Posts are parsed into entities, relationships, temporal context, and authorial stance. Vector search is hybridized with graph traversal so recall is contextual, not associative noise. 3. **Persona as a Constraint, Not a Voice Filter** Personas route retrieval and evaluation. They do not invent knowledge. Disagreements between personas are surfaced, not averaged away. 4. **Human Synthesis, Machine Recall** The system does not “synthesize insight.” It assembles grounded context so the human can. This preserves authorship while amplifying memory. 5. **Dual Publication Model** - **Human Blog**: Curated, intentional, slow. - **Persona Blog**: Autonomous, reactive, experimental (RSS-driven, news-reactive, archival reinterpretations). **Minimal Implementation Path** - Ingest blog markdown → entity extraction → lightweight knowledge graph - Add hybrid retrieval (graph + vectors) - Introduce agentic evaluations constrained to graph-backed claims - Publish two frontends from the same backend - Observe divergence over time The point is not automation. The point is **contrast**. --- ## Blog Post: After the Cat Died, I Rebuilt the Blog I didn’t rebuild my blog because of growth metrics or SEO audits. I rebuilt it because my cat died, and the old structure could no longer hold the way I think when something real breaks. Around the same time, Texas quietly rolled out new AI regulations. Nothing dramatic—just enough friction that I had to switch companies again to keep earning Meta money. It made me think about how arbitrary governance feels when it’s layered on top of systems that were never designed to be legible in the first place. I can only imagine what this looks like in the EU, filtered through German regulatory instincts—Kantian redundancy in pursuit of epistemic perfection. I grew up speaking German, not because my family was German (we aren’t), but because existing in only one language always felt intellectually suffocating. We’d visit Austria often, where my mother grew up in a convent, even though she’s more Texan than European. My father owned a plane back then. We’d fly places on weekends. He taught me how to fly. Memory is nonlinear like that. I recently started working with WeLocalize doing scout search quality rating. It’s not exciting work, but it teaches you something important: how machines are trained to interpret human intent. You learn how pages are ranked, how “needs met” is judged, how relevance is operationalized. It’s basically SEO stripped of mystique. Between that and training the Mango/Avocado project, I stay busy—but the pay is low enough that it sharpens your awareness of wasted effort. There’s a particular kind of insanity that comes from doing technically interesting work that barely keeps you fed. That tension forced a realization: my blog had become an archive, not a system. It reflected how I evolved, but it could no longer evolve with me. So I decided to rebuild it using the architecture I’ve been developing—what I call a Dynamic Persona MoE RAG system—but applied to something small and honest: my own writing. The idea is simple in principle and difficult in execution. Take every post I’ve written. Ingest them into a knowledge graph. Extract entities dynamically. Preserve citations to original documents. Then allow retrieval through a hybrid of vector similarity and graph constraints, with agentic evaluations that decide what is even allowed to be recalled. This is where people usually say “the AI synthesizes insights.” That’s wrong. The AI doesn’t synthesize anything. It reflects. It assembles stable context. The synthesis happens in the human who is finally seeing their own corpus clearly. From that foundation, the blog splits in two. One version remains mine—curated, intentional, written when I actually have something to say. The other is autonomous: a persona trained only on the same corpus, reacting to RSS feeds, news, or internal contradictions it discovers over time. Both are grounded in the same knowledge base. Both are auditable. The difference is not intelligence, but agency. This is what I meant when I once said I wanted a “stable mirror.” Not an oracle. Not a ghostwriter. A structure that lets me see what automation does to meaning when you stop pretending it’s neutral. There’s a Next.js frontend in my head for this. You ingest data. Metadata is extracted. Entities are defined as-needed, not upfront. Retrieval is constrained by what the system is allowed to notice. Outputs are forced to justify themselves against the graph. Disagreements between personas aren’t hidden—they’re the point. This architecture grew out of grief, regulation, underpaid labor, and an obsession with not lying to myself about what AI is actually good for. I’m going to apply it first to my blog because that’s the smallest place where the consequences still matter. If it works there, it works anywhere. If it fails, at least it fails honestly. --- If you want, next we can: - Turn this into a **README for the new repo** - Break the architecture into **docs/*.md planning files** - Or write the **“AI Version” first post** as a deliberate contrast