Resources
Articles, studies and demonstrators
Field notes, dated and assumed as such: the domain moves fast, and every text is a snapshot.
Needle in a haystack · Part 7 ·
The verdict: which architecture, for which haystack
Six ways to find one sentence in 33 novels, on one machine, compared.
Needle in a haystack · Part 6 ·
RLM: the model that writes its own search
The library is a variable, the model writes Python, and it never calls itself.
Needle in a haystack · Part 5 ·
Agentic search: let the model look
No index, four tools, and a model that decides where to look.
Needle in a haystack · Part 4 ·
Hybrid search: when exact words matter
What a 30-year-old keyword formula found that meaning-based search missed.
Needle in a haystack · Part 3 ·
RAG: the librarian and its catalogue
What happens when the model only reads the pages a search engine hands it?
AI Act: what technical teams need to prepare
The AI Act timeline was adjusted in 2026, but the work remains the same: inventory, classification, traceability, oversight. Compliance is decided in the architecture.
Needle in a haystack · Part 2 ·
Big context: just read everything
Can a local model find one edited fact in a 235,000-token novel?
Should you hire a head of AI?
Hiring a full-time Chief AI Officer, handing the topic to IT, or using a fractional technical leader: the criteria for choosing.
Needle in a haystack · Part 1 ·
A needle in Jules Verne
Why finding one sentence in a large corpus is still hard on your own hardware.
Why your AI proofs of concept never reach production
A proof of concept that impresses in the demo and never leaves the sandbox: five recurring causes, and how to deal with them before writing the first line of code.