Public notes with no hype: what each concept means, why it matters, and where it shows up.
The best way to learn AI is to build one small loop, make the output visible, and add a check so the system cannot only clap for itself.
This page is the map behind the public pages. If a card says tool calling, RAG, RBAC, AML, false positive, or observability, this page explains the idea without making you read the whole archive.
Start with a question, learn the term, open the risk review overview, then check the numbers. Each page is a different doorway into the same work.
These are the plain notes I would hand someone before the work gets technical: money movement, pools, lending, privacy, and the risk that follows each layer.
Why finance keeps turning into software, and where the real risk moves.
Aave lendingCollateral, rates, and why on-chain credit behaves like a live risk machine.
Uniswap poolsThe simple swap button, the pool underneath, and the risk liquidity carries.
DeFi basicsA first map for lending, borrowing, trading, and custody without the usual rails.
Zero-knowledge privacyA privacy idea worth understanding before AI and finance get too close.
Every hard concept should become one normal reading path: name the thing, show what appears on screen, explain what is happening underneath, point to the record, then name the limit.
A wallet row is a good example. The page shows a score and a note. Underneath, rules are checking patterns. The record shows which rule fired. The limit says the row is not a legal decision, not a KYC approval, and not permission to trade.
That is the goal of this knowledge base. Not more jargon. Better handles for the same system.
AI work is not only prompting. It is retrieval, evals, traces, access checks, deployment, and enough context that another person can follow the path.
Why it matters now: People understand technical work faster when they can inspect a working page instead of only reading claims.
Learn here: Use the Systems and Reports pages to move from a working page to the numbers, limits, and next question.
Question to ask: Can someone understand the path without reading the full project archive?
Follow this layer ->Why it matters now: RAG is only useful when retrieval is traceable, cited, and evaluated against expected answers.
Learn here: Study RAG, embeddings, audit logs, and observability together instead of as separate words.
Question to ask: How do AI answers cite sources instead of only sounding confident?
Follow this layer ->Why it matters now: AI teams need to know what changed, what failed, and whether a model or agent got better or worse.
Learn here: Use the Reports page to connect benchmark numbers, limitations, and system health.
Question to ask: How do teams know an AI workflow improved instead of only changed?
Follow this layer ->Why it matters now: Regulated workflows need role boundaries, approval gates, trace events, and reviewable tool use.
Learn here: Connect RBAC, audit logs, tool calling, MCP, and approval gates as one rule pattern.
Question to ask: What should an AI agent be allowed to do by itself?
Follow this layer ->Why it matters now: A demonstration becomes stronger when it can be packaged, health-checked, monitored, and redeployed predictably.
Learn here: Use Docker and Kubernetes as the path from local demonstration to reliable service behavior.
Question to ask: What makes an AI demonstration repeatable, observable, and easier to trust?
Follow this layer ->Meaning: Software that performs tasks that normally require reasoning, pattern recognition, language understanding, prediction, or decision support.
Why it matters: AI is the umbrella. The useful question is what task it helps, what evidence it uses, and where a human still makes the final call.
Shows up in: Bionic Banker / Knowledge Base / Systems
Large Language Model. The kind of AI that reads and generates text.
Meaning: A language model trained to read and generate text. It can summarize, classify, draft, reason over context, and call tools when connected to a workflow.
Why it matters: LLMs are useful when they are grounded in sources, tests, citations, and approval gates. Without that, they can sound confident while being wrong.
Shows up in: Knowledge Base / Public learning notes
Vision Language Model. An AI that reads images and text together.
Meaning: A vision-language model that can reason over images and text together.
Why it matters: VLMs are useful for screenshots, chart review, UI QA, document inspection, and visual review workflows.
Shows up in: Visual review / Knowledge Base
An AI that takes actions in a loop, not just answers a single question.
Meaning: An AI workflow that can observe state, choose a next step, use tools, write outputs, and wait for human approval when the action matters.
Why it matters: The agent is not just the model. The agent is the loop around the model: memory, tools, rules, tests, logs, and gates.
Shows up in: Agent notes / Heartbeat notes / Systems
An AI that can run real functions, not just generate text.
Meaning: A pattern where a model asks to use a defined tool, such as search, a database query, a file read, or a calculation, instead of guessing.
Why it matters: Tool calling turns a model from a text generator into a controlled worker. The boundary is the tool contract.
Shows up in: Knowledge Base / Agent rules
A toolkit for connecting AI models to real data sources.
Meaning: A developer framework for connecting language models to tools, prompts, memory, retrieval, and workflow steps.
Why it matters: LangChain is one way to structure AI apps. The value is not name-dropping it. The value is showing tool boundaries and testable behavior.
Shows up in: Knowledge Base
Meaning: A graph-based framework for building stateful agent workflows where steps can branch, loop, pause, and resume.
Why it matters: It maps well to real systems because real work is not always one straight chain. There are retries, approvals, validators, and fallbacks.
Shows up in: Knowledge Base / Agent workflow notes
Retrieval-Augmented Generation. An AI that looks things up before answering.
Meaning: Retrieval augmented generation. The model retrieves relevant source material first, then writes an answer using that material.
Why it matters: RAG is how you reduce guessing. The question becomes: what did the model retrieve, can we cite it, and was the source trustworthy?
Shows up in: Visible evidence notes / Source-linked explanations / Question pages
Meaning: Numeric representations of text, images, or other data that help software compare meaning instead of exact words.
Why it matters: Embeddings make search smarter. They help find related ideas even when the wording is different.
Shows up in: Retrieval / Knowledge search / Semantic search
Meaning: Model Context Protocol. A standard pattern for letting AI systems connect to tools and data sources through defined interfaces.
Why it matters: MCP matters because agents need clean boundaries. A tool should say what it can do, what inputs it accepts, and what output it returns.
Shows up in: Knowledge Base / Tool-boundary design
Role-Based Access Control. Who can see what, and why it matters in finance.
Meaning: Role-based access control. A way to decide who or what is allowed to perform each action.
Why it matters: RBAC is essential for regulated AI because not every agent should be allowed to do every action.
Shows up in: Knowledge Base / Agent rules
A time-stamped record of who did what. Common in finance and compliance.
Meaning: A record of what happened, when it happened, what tool or person did it, and what evidence supported it.
Why it matters: If a system cannot explain its own actions later, it is not ready for serious finance, healthcare, or compliance work.
Shows up in: Wallet Risk Assessment / Agent notes / Agent rules
The ability to see what is happening inside a system while it runs.
Meaning: The ability to see whether a system is alive, what it is doing, what failed, and what changed over time.
Why it matters: Invisible automation is risky. Observable automation can be trusted, debugged, and improved.
Shows up in: Heartbeat notes / Knowledge Base
Meaning: A small status signal that shows a system woke up, ran, and wrote its latest state.
Why it matters: A heartbeat does not prove a system is perfect. It proves the loop is still alive and observable.
Shows up in: Heartbeat notes
Anti-Money Laundering. The rules that flag suspicious financial patterns.
Meaning: Anti-money laundering. The work of detecting, investigating, and preventing financial activity that may be tied to crime or sanctions risk.
Why it matters: AML is where finance, data, rules, and judgment meet. Good AML systems explain why a transaction looks risky.
Shows up in: Wallet Risk Assessment / Wallet Risk Notes / Wallet Risk Check
A flag that fires when it should not. The analyst still has to clear it.
Meaning: A system flags something as risky, but after review it is not actually risky.
Why it matters: False positives matter because they waste review time. In compliance, accuracy is not only detection. It is also how much noise the system creates.
Shows up in: Wallet Risk Assessment
Ethereum Virtual Machine. The engine that runs smart contracts on Ethereum.
Meaning: Ethereum Virtual Machine. The execution environment used by Ethereum and many compatible blockchain networks.
Why it matters: EVM compatibility lets one wallet-risk tool support many chains with similar address and transaction patterns.
Shows up in: Wallet Risk Check
A system that keeps software services running reliably at scale.
Meaning: A system for running, scaling, and managing containers across machines.
Why it matters: Kubernetes is useful when a project needs deployment discipline, service health, scaling, and reproducible operations.
Shows up in: Deployment learning
A way to package software so it runs the same way on any machine.
Meaning: A way to package an app and its dependencies into a container that can run consistently across environments.
Why it matters: Docker makes projects easier to test, share, and deploy because the environment is part of the package.
Shows up in: Deployment learning / Agent services
Meaning: A graphics processing unit. In AI, GPUs are used to run many calculations in parallel.
Why it matters: GPUs matter for model training, image models, local inference, and workloads where parallel math is the bottleneck.
Shows up in: Hardware learning