Evidence before scale
We test important assumptions against real work before we expand scope.
The studio
Move from a useful question to a working system.
Sinumo works with software, product, and executive teams to choose valuable AI work, build it into real operations, and transfer the capability to the people who own the outcome.
We also build products. Repeated problems in client work can reveal a broader need, but client information and client-specific systems stay separate from our product development.
01 Decide with evidence
02 Build for operations
03 Transfer real ownership
01 Decide
Choose the work worth doing.
AI strategy should produce decisions, not a catalogue of possibilities. We connect business value to technical evidence, then define the smallest credible path to production.
What we do
- Map decisions and workflows before proposing technology.
- Assess data, integration, security, and operating constraints.
- Prototype the uncertain parts with real inputs and users.
- Compare build, buy, adapt, and defer options on the same criteria.
What leaves with you
- Opportunity portfolio
- Evidence-backed priorities
- Target architecture
- Delivery and evaluation plan
02 Build
Make the system dependable.
A useful model is one component. We build the context, tools, permissions, evaluations, and recovery paths that turn it into a production system.
What we do
- Design agentic workflows with explicit state and stop conditions.
- Connect product, engineering, knowledge, and operating systems.
- Build retrieval, knowledge graphs, tool contracts, and model routing.
- Add evaluation, observability, approval, and rollback from the start.
What leaves with you
- Working production software
- Versioned evaluation suite
- Operating controls
- Runbooks and decision records
03 Transfer
Leave the team stronger.
The work is not complete when a vendor can run it. We work beside the people who will operate, review, and extend the system after the engagement.
What we do
- Pair with engineers through architecture, implementation, and review.
- Give product leaders practical ways to specify and evaluate AI behavior.
- Create policies for model choice, tool authority, and human checkpoints.
- Train teams with their own system, cases, and failure modes.
What leaves with you
- Owned code and infrastructure
- Architecture and policy guides
- Team-specific training
- A measured improvement backlog
Engagement model
A staged path with useful exits.
Each phase must produce evidence or an asset that remains useful if the next phase does not proceed. Scope grows only when the evidence supports it.
- 01
Discover
Observe the work, interview the people, and locate the decisions that matter.
- 02
Frame
Define the outcome, constraints, baseline, and evidence needed to proceed.
- 03
Prove
Test the hard assumptions with a bounded prototype and representative cases.
- 04
Ship
Integrate the system, establish controls, and release it through measured stages.
- 05
Transfer
Move knowledge, authority, and the improvement loop into the team that owns it.
How we work
Clear constraints. Inspectable progress.
Controls at the boundary
Permissions, verification, and approvals live where actions happen—not only in prompts.
One accountable outcome
Each engagement has a result that an engineer, product leader, and executive can inspect.
Built to be owned
Architecture, code, and operating knowledge remain legible to the team that receives them.
The team
The people behind the studio.
Practitioners in AI strategy, AI engineering, data, platform systems, and UX/UI who partner directly with client teams from initial opportunity through handover.
Dario Farzati
Founder, AI Strategy
Adrian Caneva
Data
Founder, E2E2
Leandro Barbagallo
Backends & Platform
Founder, Fulgurion Systems
Wojciech Łęcki
AI Engineer
Priscila Gacio
UX/UI
Independent Consultant
Product work
We build for recurring problems, too.
Consulting shows us where teams repeatedly lose time, context, or control. When the pattern is broad and durable, we may turn the shared problem into a product.
Product work follows the same standard as client work: a defined user, a measurable job, production controls, and evidence that the system is useful outside a demonstration.