AI pipelines on your data
I turn the pile into one window.
Ask in plain words, get an exact answer with its source. Just work. I build agentic pipelines and RAG databases on your documents: AI agents and the harness they run on, turnkey.
Master chat
Update the court base and check every law edition.
Split into 3 tasks, agents assigned. memory: project log + handoff loaded
Source attached for every change.
Draft
Parse new rulingsagent: claude
Check
Wrong valuesagent: codex
Act
Law index updatedagent: opencode
The pile
Every organisation is drowning in paper.
Rulings, laws, contracts, mail. People search, instead of working.
- Rulings
- Laws
- Contracts
The window: ask in plain words
Can my landlord raise the rent twice a year?
No. A rent increase is allowed once per 12 months, with written notice at least two months in advance.
Source, p. 4
Illustrative example. Every answer comes with its source; no source, no answer.
How it works
Underneath: millions of documents, indexed, linked, checked.
Data that AI agents can finally work with.
Your documents
Rulings, laws, manuals, mail: indexed, linked into a graph, checked. No source, no answer.
Agents
AI agents take the data and do the work: draft, check, act. Claude Code, Codex and opencode.
Tools
Calculators, verifiers, search, code. Agents do the maths with your numbers and catch wrong values.
The harness
A master chat, a board and shared memory, so no context is ever lost. About ENKI
- For law
- For engineers
- For people
- For science
Overview
One pattern, every industry.
Different industries, one pattern: your documents, one window, no more piles.
What I’ve built
That’s what I do. Not on slides.
I’ve built this window again and again. For law, for engineers, for people, for science. Take a look.
Law
-
caselaw.help United States
Three and a half centuries of US case law.
- 9.7Mcourt decisions
- 1.9Mstatute sections
- 1658–2026years covered
-
rechtzoek.online Netherlands
The Netherlands in one graph.
- 3.6Mrulings, plus parliament
- 3.147Mnode graph
-
pravo.help Ukraine
Every edition of every Ukrainian law.
- 23Mcourt decisions
- 287Klaws, any edition
Infrastructure for AI agents: agents draft, check and act on these bases.
Engineering
-
Books that do the maths
Formulas from technical books, calculated with your numbers. Agentic AI tools for R&D and engineering bureaus.
-
Parts assistant
- 686Kparts
-
Prompt to Verilog to FPGA
Type “blink LED at 1 Hz”, get Verilog, the board blinks. 3, 2, 1, lift-off.
-
Photos of pages to checked data
- 50wrong values caught
People
-
AI CRM with human proof
The letter writes itself from the profile; a person approves it.
-
refuge.help Live
Rare, uncovered languages. Real time. Offline.
- 1600+languages, speech recognition
- 400+languages, translation
-
voiceresume.online
Just talk, get a CV.
Research & code
-
Research pipeline
Hypothesis, knowledge graph, experiment data, sixteen layers of tools: calculator, code, search, charts; critique and synthesis in a loop. One hypothesis becomes three. Faster science.
-
PICO Research memory
Papers cut into fragments for the agent’s memory.
- 6.3Mchunks
-
Markets Research, paper only
Strategy research on OKX, Binance and Hyperliquid data.
The harness
ENKI, my harness for agents on huge bases.
Graphs and nine memory layers, so no context is ever lost.
Agents on Chromium
The Chromium source tree becomes a graph; the agent walks it to the right file. When its context window fills up and resets, the facts stay in memory.
- 507Kfiles
- 880Knode graph
The same harness runs over other bases:
- 10M knowledge base
- Laws & accounting
- Research tools
Case study, offline
Nine memory layers
- Master map
- Index
- Notes
- Project rules
- Project logs
- Handoff
- Cross-project log
- Daily logs
- Semantic search over all layers
- Kept in git
- Synced every 30 min
- Cold backup
Real work
AI pipelines, built on your data.
Different industries, one pattern: your documents, one window, no more piles.