[ 01 ]
Evidence over opinion
Recommendations come with the measurement that produced them. If we say a path fails, we can show you the trace. If we say an AI agent is not ready, we can show you which intents it misses and how often.
Why we exist
Two large things are happening to enterprise voice at the same time, and they are usually handled by two different sets of people who do not talk to each other.
On one side, the copper network underneath the last forty years of business telephony is being retired on a compressed schedule. On the other, organizations are putting conversational and agentic AI in front of customers faster than they can evaluate it.
The first problem needs someone who can read a DS1 alarm log. The second needs someone who can design an evaluation harness for a language model. Very few teams have both, and the gap between them is where projects quietly go wrong: a cutover that strands an elevator phone, or an AI agent that demos beautifully and fails the twelfth path a real customer takes.
KuberEva was built to sit in that gap.
- founded by
- Alfredo Montesinos
- experience
- 25 years
- clients
- SMB → Fortune 500
- most recently
- Principal Architect, Cyara
- engagement model
- Advisory & architecture
- stance
- Vendor-neutral
Principal
Alfredo Montesinos
Twenty-five years designing contact center, conversational AI, and customer experience systems for organizations ranging from small businesses to the Fortune 500.
Most recently Principal Architect at Cyara, where I benchmarked NLU quality across large language models including GPT-4, and evaluated conversational AI agents for production readiness. That work is the direct ancestor of how this practice approaches AI: not "does it answer well in a demo," but "what happens on the paths nobody scripted."
The earlier part of that career was spent on infrastructure that is now being decommissioned — the trunk groups, dial plans, and signaling that carried the calls before any of this was cloud. I do not treat that as a museum exhibit. It is running, today, underneath a great many organizations that are about to receive a retirement notice.
LinkedIn ↗Areas of direct experience
| Domain | Depth |
|---|---|
| Contact center architecture | Design and delivery, SMB to Fortune 500 |
| Conversational AI evaluation | Production-readiness assessment of IVAs and voice bots |
| NLU / LLM benchmarking | Comparative quality measurement across models including GPT-4 |
| CX assurance & testing | Automated voice and digital channel testing |
The team
A bench that spans four decades of telephony
We work as a small senior team rather than a pyramid. Nobody is learning on your project. Collectively, the practice has hands-on experience across every layer below — which matters because a migration touches all of them at once.
Six eras, still overlapping
1980 → 2030
| Era | What we ran | Why it still matters |
|---|---|---|
| Analog | POTS · loop start · ground start · Centrex · key systems | Elevator phones, fire panels, and POS terminals are still on these lines and are the hardest part of any retirement. |
| TDM | T1 / DS1 · fractional T1 · E1 · channel banks · DACS | Circuit inventories are almost always wrong. Reading the actual provisioning is how you find what you'd otherwise strand. |
| Digital signaling | ISDN PRI & BRI · 23B+D · SS7 · CAS · robbed-bit | Feature behavior — caller ID, DNIS, redirect, transfer — is defined here, and has to be reproduced in SIP. |
| IP voice | SIP · RTP · SBCs · SIP trunking · SD-WAN | The destination architecture, and where most of the quality and security problems land. |
| Cloud CX | CCaaS platforms · routing · WFM · CRM integration | Where the contact center actually lives now, and what a migration is usually funded to reach. |
| AI | IVAs · voice bots · NLU · LLM agents · evaluation harnesses | What is being layered on top, frequently before anyone has defined what "working" means. |
A note on scope
We are an advisory and architecture practice. We are not a carrier, a reseller, or a
managed service provider, and we do not take vendor commissions. That is a deliberate
constraint: it is the only way a platform recommendation means anything.
How we work
Four commitments that shape every engagement. They are also the fastest way to tell whether we are a fit for you.
[ 02 ]
Inventory before architecture
Nearly every migration problem we have seen traces back to a circuit, a line, or a dependency nobody knew existed. We find those first, before anyone draws a target state.
[ 03 ]
Your team owns the result
We design for handover. Runbooks, decision records, and measurement your staff can run without us. An engagement that creates a dependency on us has failed.
[ 04 ]
We will tell you no
If a use case is not ready for AI, if a platform is wrong for you, or if the work is outside what we do well, we say so. Being easy to hire is not worth being wrong about.
What we don't do
Stated plainly, so nobody wastes a meeting.
- We do not sell circuits, licenses, or hardware, and we take no vendor commissions.
- We do not provide 24×7 managed operations or NOC services.
- We do not staff-augment development teams by the seat.
- We do not certify compliance. We architect toward it and work alongside your QSA, counsel, or auditor.
Where we fit best
The engagements that go well tend to share a shape.
- A carrier notice has landed, or is about to, and the true circuit inventory is unclear.
- A CCaaS decision is being made and the internal debate has stalled on vendor claims.
- An AI agent is being pushed toward launch and nobody has defined what "ready" means.
- A migration already went badly and someone needs to find out why.