The Blog

My exploration of harness engineering, and the primitives that allow the agents to do work on loop even when designing harnesses

I wanted to see how far I could push autonomous code generation. Not by writing better prompts, but by building a system where agents could implement, verify, and fix their own work without me watching. A DOCX editor built from behavioral specs and pixel diffs was the testbed.

How come we have advanved agent tracing products with minimal setup, but prompt optimization itself needing rewriting your product in new frameworks in order to work. My attempt on showing that it does not need to be the case

How to create AI systems that adapt and improve their performance over time by learning from past interactions. Going over the MemGPT and Letta AI frameworks approach on bulding agents with memory systems that evolve through conversations.

Learning and building automated evaluation system with Ragas. Can LLMs automatically evaluate RAG systems? Are human labels always needed?

Exploring the practical benefits of the Matryoshka Representation Learning technique through my own experiment. Why some models can be shrunk without "losings" performance, how it works, and does it always work?

Explaining and building real use case for Hypothetical Document Embeddings (HyDE) to showcase that hallucinations are not always bad

Why LLMs are sensitive to their early tokens, and where prompting techniques like Chain-of-Thought and Tree-of-Thought get their power. This is my attempt to understand why early tokens matter, what ToT does behind the scenes, and whether it’s worth using outside of toy projects.

Building GEPA prompt optimizer from scractch to optimize compound AI system prompts using DSPy framework.

Real world use case for building RAG and learning from the building. Building fully functional travel itinerary planner with AI agents and RAG.