Evals as Tests, Part 2: What Happens When the Prompt Injection Is Inside a PDF?
Attest can now look for prompt injections in PDFs your agent reads!
read →writing.
thoughts on engineering, systems, and things i find interesting.
#AI Engineering#RAG
Beyond Vector Similarity: Ranking Authority, Currency, and Relevance in Legal RAG
How Vera combines vector similarity with legal status, temporal context, lexical matching, source authority, and document-type-specific ranking to retrieve current law without losing historical coverage.
read → #AiEnginnering
Evals as Tests: Grading Your Agent with attest
After building a code-review agent, the next question is whether it keeps behaving. This post turns that into a test suite: a spec of expected behavior, graded automatically by attest, that fails when the agent regresses.
read → #AIEngineering
Building a Basic Code-Review Agent with LangChain
A practical walkthrough of building a focused, tool-using code-review agent with LangChain's create_agent and structured output — the model, the tools, the response schema, and the system prompt that holds the agent to its role.
read → #AIEngineering
Thought, Action, Observation: The Loop Behind Every Useful AI Agent
An AI that only reasons hallucinates; one that only acts is aimless. The ReAct loop — Thought → Action → Observation — braids them together, grounding every step in what actually happened.
read → #Machine Learning#Vectors
Vector Analysis and Machine Learning
A vector — an ordered list of numbers carrying both magnitude and direction — can represent anything from the pixels of an image to the semantic meaning of a word.
read →