Medium

#hope

#hope

·

Jul 13

Taxation without Transportation: The MA — NH Conundrum

The traffic between MA and NH has only gotten worse over time

Taxation without Transportation: The MA — NH Conundrum

#hope

·

Jul 8

The Last Mile: Recency, Quality Filters, Soft-Delete, and Context Assembly

You’ve retrieved and fused a strong set of candidates (Article 7). You are not done. The gap between “relevant” and “the right thing to…

The Last Mile: Recency, Quality Filters, Soft-Delete, and Context Assembly

#hope

·

Jul 7

Hybrid Search and Reciprocal Rank Fusion: Why one retriever isn’t enough

After all the import-time work — enriched chunks, atomic facts, a timeline — you finally run a query. The temptation is to do one semantic…

Hybrid Search and Reciprocal Rank Fusion: Why one retriever isn’t enough

#hope

·

Jul 6

Give your Knowledge a Timeline: Structured Events and Causality

Some questions are fundamentally about time, and embeddings are hopeless at them. “What changed right before the October outage?” “Is this…

Give your Knowledge a Timeline: Structured Events and Causality

#hope

·

Jul 3

Deduplication as a feature: Turning repetition into a relevance signal

Once you start extracting atomic facts, you immediately create a new problem: the same fact gets extracted hundreds of times. In our…

Deduplication as a feature: Turning repetition into a relevance signal

#hope

·

Jul 2

From Prose to Atoms: Extracting Structured Facts From Unstructured Text

Embeddings make whole passages findable. But often you don’t want a passage — you want a single, clean fact. “What’s the actual fix for the…

From Prose to Atoms: Extracting Structured Facts From Unstructured Text

#hope

·

Jul 2

Enrich, then embed: The trick that fixes vocabulary mismatch

This is the most important idea in the series.

Enrich, then embed: The trick that fixes vocabulary mismatch

#hope

·

Jul 1

Chunking Done Right: Normalization, sentence boundaries, and overlap

Chunking is the least glamorous part of a retrieval pipeline and the one that quietly wrecks the most systems. Get it wrong and every later…

Chunking Done Right: Normalization, sentence boundaries, and overlap

#hope

·

Jun 30

Why you shouldn’t search your documents directly with AI

Everyone’s first retrieval-augmented system looks the same: chop the documents into pieces, turn each piece into a vector, search the…

Why you shouldn’t search your documents directly with AI

#hope

·

Jun 29

Teaching software to actually read your documents

There’s a second, quieter kind of AI product that doesn’t chat with you — it reads a pile of your documents and turns them into something…

Teaching software to actually read your documents

#hope

#hope

“If you see something that’s not right, not fair, not just, do something about it. Say something. Do something.” — Rep. John Lewis

Following

Help

Status

About

Careers

Press

Blog

Store

Privacy

Rules

Terms

Text to speech