I have spent twenty-three years solving problems for other people. Technology problems, business problems, people problems. The irony is that the systems I use to manage my own life are held together with sticky notes, calendar reminders, and a memory that works well enough until it does not.
Tessera started as a question: what if I built an assistant that actually understood how I think? Not a chatbot. Not a task manager with a language model bolted on. Something that could look at the full context of my life, my decisions, my patterns, my commitments, and help me navigate the complexity without losing the nuance.
The commercial assistants do not do this. They are stateless, generic, and disposable. They do not remember that I made a commitment to a client three months ago that affects a decision I need to make today. They do not understand that my approach to technical remediation follows a specific pattern that I have refined over two decades. They do not know that when I say “handle it,” I mean something very specific depending on the context.
The Architecture Gap
I looked at every RAG architecture available. Vector similarity search over chunked documents. Fine-tuned models. Knowledge graphs. They all share the same limitation: they treat retrieval as a search problem. You ask a question, the system finds relevant chunks, and a language model synthesizes an answer.
That works for documentation lookup. It does not work for a life assistant. A life assistant needs to understand relationships between pieces of information that were never explicitly connected. It needs to know that a conversation I had with a vendor last Tuesday is relevant to a staffing decision I am making this Friday, even though no keyword connects them.
The architecture I need does not exist. So I am building it.
What Tessera Needs to Do
The requirements are deceptively simple. Tessera needs to ingest everything: emails, documents, meeting notes, technical documentation, personal reflections, decision records. It needs to understand not just what those artifacts say, but what they mean in the context of everything else it knows.
When I am facing a technical remediation scenario, a client environment that is failing, a security incident, a migration gone sideways, Tessera should be able to surface not just relevant documentation but relevant decisions. How did I handle something similar before? What worked? What did I wish I had done differently?
That is the core insight. The unit of value is not the document. It is the decision. Every artifact in my corpus is evidence of a decision made, a pattern followed, or a judgment exercised. Tessera needs to index at the decision level, not the document level.
Why This Is Different
Standard RAG retrieves text. Tessera retrieves judgment. Standard RAG matches keywords and embeddings. Tessera matches patterns and contexts. Standard RAG answers the question you asked. Tessera should answer the question you should have asked but did not think to.
This is not incremental. It requires a fundamentally different retrieval architecture. One that combines graph-based relationship mapping, vector similarity for semantic matching, and a verification layer that cross-references retrieval results against the full decision history to filter out noise.
I am calling it hybrid trimodal retrieval. It is the core innovation, and everything else in Tessera depends on it working.