Evidence-led business improvement
Improvement you can check.
We measure the process as it actually runs, build the agents that close the largest gap, then re-evaluate the same measures. The delta is the deliverable — not the deck.
Continuous Improvement, Next-generation Tooling, Evidence-Led Intelligent Systems
Sectors we have delivered in
Built where the process is
Every sector hides its waste somewhere different.
Regulated industries carry the same three costs — manual handling, rework, and evidence produced after the fact. What differs is where they sit. Fifteen years across these sectors is how we know where to point the instrument first.
Care plans that write their own evidence.
Clinical and operational data spread across hundreds of systems, with audit evidence assembled by hand at the worst possible moment. We connect it into a queryable graph and put agents on the reading, so compliance evidence falls out of normal work instead of a scramble.
Submission work that stops eating the week.
Lending, broking and advisory run on document assembly, re-keying and chasing. It is the most measurable waste in the business and the least defended. We instrument it, then put a bounded agent pipeline behind a single human approval gate.
Assurance evidence as a by-product.
Public sector improvement lives or dies on whether the change survives audit. Every agent action is logged, attributable and replayable against the control framework you already report on — so the improvement and the evidence arrive together.
Deterministic work only, and only behind a gate.
Advisory firms sell judgement, so agents must never appear to supply it. We build systems where the model does classification, extraction and diffing, and a named human owns every outbound artefact. The improvement is throughput without a reputation risk.
Systems in production
What we hand over.
Not recommendations. Running systems, each one built because a measurement said it should exist, each one still producing evidence after handover. All of these are live — click through and use them.
Live demo
Graph Query Agent
Operations questions become reviewed, parameterised Cypher over a knowledge graph. Every value traces to the query and node IDs behind it. Writes sit behind an approval gate.
graphagent.eqr.vc →
Live demo
Agent delivery pipeline
Point it at a repository and describe a change. It maps the codebase into a graph, plans, edits, checks its own work, then opens a pull request. Every run costed.
design-graph.totallywild.ai →
Live demo
Buyer Outreach
Sell-side M&A pipeline where every send, spend and counter-signature waits behind an advisor gate. The model classifies and diffs. It never writes the outreach.
buyer-outreach.eqr.vc →
Live demo
H4Graph
Ask a corpus a question, get an answer where every value carries its citation — passage, DOI, date. An append-only lineage store means it cannot emit an untraceable number.
h4graph.vercel.app →
Live product
Brainstorm Board
A thinking workspace where notes stay plain Markdown files, the link graph builds itself from wikilinks, and a described architecture becomes an editable draw.io diagram rather than a picture of one.
brainstorm.cintelis.ai →
Live demo
Agent observability
Cost, health and latency for everything we deploy, behind single sign-on with no public IP and no inbound firewall rule. Alerts provisioned as code. About $55 a month.
agent-observability-demo.pages.dev →
Cycle report · re-evaluation
Submission prep stopped being the constraint.
A non-bank lender in New Zealand was losing most of an adviser's week to document assembly and re-keying. We took the baseline in week one, deployed a five-agent pipeline behind a single human approval gate, and re-evaluated the same measures against the same source systems.
▼ 88% per submission · same measures, same sources · verified against week-one baseline
Client named only with written consent. Every figure published here names its baseline and its measurement method.
88%
Less admin effort per submission
5
Agents in the pipeline, one approval gate
0
Change to who holds approval authority
13
Platforms shipped and live
553K
Graph nodes in production
15+
Years across regulated sectors
How the work runs
Diagnose, then build. Never the other way round.
Phase one
Take the reading
We instrument the process as it actually runs, not as the process map says it does. Cycle time, touch count, rework rate, cost to serve, where work waits. No workshop producing a wish list.
Output: a baseline with sources
Phase two
Build the instrument
Agents and tooling deployed into the real process, sized to the largest measured gap. Access control, logging and approval gates are part of what ships, not a later phase.
Output: a system in production
Phase three
Re-evaluate it
Same measures, same sources, after the change. If the number did not move we say so. Then the next largest gap becomes the next baseline, and the measurement keeps running once we leave.
Output: a verified delta
Start with the assessment. Keep it either way.
Fixed scope, fixed fee. We instrument one process end to end and hand back a baseline that is yours, whether or not you build anything with us.
Get in touch
Tell us where the week goes.
Name the process that eats the most time and we will tell you what an assessment of it looks like — fixed scope, fixed fee, no pressure and no lock-in.
- Email[email protected]
- Phone1300 786 040
- LinkedInCintelis AI
- BasedBrisbane, Queensland