Duna Solutions
AI implementation consultancy

We show you what AI can do for your business — by building it.

Duna Solutions builds specialized AI systems for small and mid-sized companies that don't have an AI team. First contact is a working demonstration on your data — then we scope, build, deliver, and maintain it.

Book a discovery call See a sample artifact
Scroll
0
automated tests across our delivered systems
0
modules in our largest client cockpit
0
AI models cross-checking every concept
0
sub-agents in one architecture review council
EN·中文
bilingual executive delivery
The artifact

What lands in your inbox before any contract.

Every engagement starts with a demonstration built for your business. These are the three panels of a typical Duna artifact — rendered here with illustrative sample data.

duna · artifact preview

Category trend radar

12-week signal momentum · beverage sample category
Osmanthus oolong
+86
Sparkling cold brew
+71
Coconut cloud top
+64
Sea-salt guava
+52
Charcoal jelly
−23
Sample data — illustrative

Concept brief · RTD-07

Generated by 5 models · ranked against your production line

Osmanthus Oolong Sparkler

Low sugarClean labelCold chain ready21-day shelf life
Category momentum8.6
Fits your base stock9.1
Process capability7.4
Channel match7.9
8.2
Overall fit
Sample concept — illustrative

Evidence chain · RTD-07

Attached by code, not by the model — it cannot invent a citation
Social signals clusteredshort-video and review posts, deduplicated214 posts
Competitor gap confirmedno incumbent SKU in the niche at target price3 surveys
Internal capability matchedyour equipment and certifications, from your own docs2 lines
Concept generated & cross-checkedfive models propose, score, and challenge5 models
Human decision recordedyour team's approval is signed and persistssigned
Sample chain — illustrative
Process

Lead with a demonstration, not a pitch.

Most consultancies open with a deck. We open with a working artifact built on your market and your data — and only then talk scope.

Step 01 — Demonstrate

The artifact

Costs you a conversation

A personalized proof of concept: a short insight report and a small working demo built on sample data from your business — before any contract is signed.

Step 02 — Scope

The discovery call

We walk through what the demonstration found, agree on what the full system should decide for you, and fix scope, timeline, and acceptance criteria in writing.

Step 03 — Build

The system

A staged build with automated tests at every layer. You sign off at delivery acceptance, and an observation window after launch proves the system in daily use before the engagement continues.

Step 04 — Retain

The partnership

We maintain, monitor, and extend the system on retainer. Capacity governs how many clients we take — we never sell more than we can deliver well.

Work

Evidence, not adjectives.

Real systems, real numbers — claims limited to what we can show.

Client // FStea 斐素 — endorsed case study

New Product Opportunity Cockpit

A bilingual nine-module decision cockpit for a Shanghai beverage maker with its own brand and a private-label manufacturing arm. It fuses social, retail, and internal signals into CEO-facing product briefs: trend radar, competitive tracking, multi-model concept generation, and fit-scoring against the company's own production capabilities — so recommendations are things FStea can actually make and sell.

  • 0cockpit modules, English and Chinese throughout
  • 0AI models generating and cross-checking concepts
  • 0automated tests across the analysis engine
  • 证据链every output carries a code-enforced evidence chain
Lab // Own product

Trus — generative personal software

From the same founding team: a platform that generates working software from a description. Describe the tool you need, review the blueprint an agent proposes, and watch it get built and rendered live. It's where we pressure-test the build systems Duna clients benefit from.

  • 0backend tests passing
  • 0frontend tests passing
  • 9.4/10automated design-gate score on generated apps
In development

Project Bev Coming soon

A drink-idea sourcing engine for beverage and F&B manufacturers: short-video and retail-shelf signals in, ranked product recommendations out. The next cartridge in the engine.

  • —your industry could be the one after
"The deliverable is decisions, not modules — recommendations an executive will stake money and shelf space on."
How we measure our own work
01Niche to amortize 02Personalize or perish 03Quality is a first impression 04Manual before machine 05Deliver above all
Systems we build

The engine is proven. The cartridge is yours.

Our delivery engine — target, demonstrate, build, retain — stays the same. What we build with it is shaped to your industry. Examples of the shapes it takes:

Opportunity cockpits

Decision dashboards that fuse market, competitor, and internal signals into ranked, evidence-linked recommendations.

Cartridge · market intelligence

Signal collector fleets

Always-on pipelines reading social platforms, retail shelves, and trade sources — deduplicated, translated, and scored.

Cartridge · data collection

Concept generators

Multi-model idea engines that propose, score, and challenge new product concepts against what you can actually produce.

Cartridge · generation

Ops copilots

Assistants wired into your real workflows — quoting, scheduling, reporting — with human sign-off where it matters.

Cartridge · operations

Report engines

Executive briefs generated on schedule — bilingual if your team is — with every figure traced to its source.

Cartridge · reporting

Your industry's cartridge

The placeholder we're most interested in. If your business runs on decisions buried in messy signals, it's a candidate.

Cartridge · next
Method

Auditable AI, or nothing.

Executives shouldn't have to trust a model's judgment. Our systems are built so they don't have to.

M-01

Evidence chains

Every recommendation carries its sources — which posts, which competitor gap, which internal capability — attached by code, not by the model, so the system structurally cannot invent a citation.

M-02

Human-in-the-loop

Approve, reject, rename, merge, re-rank — every judgment is signed and persists, re-applying on every future run. Your team's decisions become the system's quality signal.

M-03

Decisions, not modules

The deliverable isn't software features. It's decisions an executive will stake money and shelf space on — with the reasoning attached.

M-04

Tests before claims

Every layer ships with an automated suite, and we prove a bug exists before claiming the fix. Features count as done only after hands-on verification in the real product.

Founders

Built by people who ship agentic AI for a living.

Co-founder

Diego Sanchez

Hands-on agentic-AI build experience at Amazon. Systems and product engineering: analysis engines, evaluation harnesses, and the tooling that keeps autonomous builds honest.

Co-founder

Janus Tsen

Stanford; agentic AI at Morgan Stanley Wealth Management. Platform and intelligence engineering: data pipelines, shared infrastructure, and client delivery.

FAQ

The questions every prospect asks.

What does the first step cost?

The demonstration costs you a conversation. We build the initial artifact — an insight report and a small working demo on sample data — before any contract, because we'd rather show than pitch. Build and retainer pricing is scoped on the discovery call, sized to the system.

How long does a build take?

A typical full build lands in about a quarter: staged delivery with acceptance criteria you sign off on, followed by an observation window where the system proves itself in daily use before the retainer begins.

We don't have a data team. Is that a problem?

It's the point. We build for companies without in-house AI teams — the retainer covers maintenance, monitoring, and extension, so the system keeps working without you hiring for it.

How do we know the AI isn't making things up?

Structurally. Citations are attached by code, not generated by the model, so a claim without a source cannot exist in the system. Your team can inspect, override, and re-rank anything — and those decisions persist.

What industries do you work with?

The engine is industry-agnostic; the deepest demonstrated domain is beverage and consumer goods manufacturing, including bilingual delivery for teams operating in China. If your business runs on decisions buried in messy signals, it's a fit worth a call.

Who owns what we build?

The system is built for your business and runs for your business. Ownership and licensing terms are fixed in writing at scoping — no surprises at delivery.

Contact

Tell us your business.
We'll show you the system.

Engagements start with a demonstration built for you, not a pitch deck. Email us with a sentence about your company and we'll take it from there.

hello@dunastrategies.com