EVO Data Advisor
A senior data consultant that interviews you, challenges the premise and designs the architecture your scenario actually needs. It sells no database, takes no commission from any vendor and executes nothing in your environment: it hands over the verified assessment and the package ready for whoever runs it.
Paste your schema and get the most serious findings in about two minutes. No account.
Now: alternatives compared against weighted criteria, cost computed per tool, assumptions on record.
We want to move everything to a data warehouse. Which one do you recommend?
Before choosing, I need two numbers I do not have yet: your active volume and how many people query per day. You mentioned 40 GB and 25 BI users. At that size, a warehouse usually costs more than it solves.
I will record that as an assumption, with the condition that would invalidate it: if active volume passes 1 TB, the math changes.
Makes sense. Compare the two options then.
Requirements captured
Assumption on record: "active volume stays below 1 TB". Re-checked on every read of the environment.
Who it is for
And you would rather the comparison did not come from the sales deck of one of the candidates. The matrix scores both, and states what it costs to leave each one.
It grew by accretion, nobody remembers why that column exists and the bill climbs every month. The assessment ranks findings by severity, with the evidence inside your own artefact.
The decision goes to a committee that was not in the room. The same content comes back recalibrated for the reader you name, as its own document.
It is not for someone who already decided and just wants to ship faster. If the platform is settled and the work now is building pipelines inside it, that vendor own agent does it better than we would. EVODA exists for the question that comes before.
What we believe
Every data agent on the market belongs to someone who sells the answer. Advice that decides architecture has to come from someone who gains nothing from the choice.
The whole category got very good at executing inside its own house. The data assistants from the large vendors build pipelines, tune queries and refactor models, each one within its own platform. It is excellent engineering, and it answers "how do I do this in here".
Nobody moved toward answering "is in here the right place at all?". That is coherent: an agent owned by a vendor has no incentive to recommend leaving it.
Vendor lock-in is a declared concern for 94% of organisations surveyed in the Parallels cloud survey, covering 540 IT professionals in the United States, United Kingdom and Germany.
Answers fast and well, as long as the question fits inside the product that employs it. The platform choice was made before it entered the room.
Arrives before the choice, compares options against weighted criteria and states the cost of leaving each one. Then hands the package to whoever executes.
Draws schemas carefully and reverse-engineers well. Does not pick technology, does not compute cost and does not remember last quarter decision.
Remembers what was decided, why it was decided and the assumption that held it up. When the assumption breaks, it points at the affected decisions.
The honest promise
We do not promiseTo get the architecture right
We promiseTo show how each recommendation was verified, and where it could be wrong
A computed confidence seal, a recorded red team and a second opinion from a different vendor model, all published inside the deliverable.
We do not promiseAn exact cost estimate
We promiseThe arithmetic in the open, with a source for every price, and our own deviation published
Cost is computed by tooling, never from memory, and later reconciled against your real invoice.
We do not promiseTo replace your data team
We promiseTo give your team the assessment it would take weeks to assemble
The methodology canon is curated and versioned, and the interview is led by something that knows what to ask.
We do not promiseTo execute in your environment
We promiseTo hand over the package ready for whoever executes
Validated DDL, verification steps, rollback and the fenced prompt for the agent or the person who will run it.
After the engagement closes, EVODA keeps watching: it takes in the report from whoever executed, compares the real invoice against what it projected, and warns you when an assumption that held up a decision stops being true.
How it works
EVODA never connects to your production database to change anything. What comes out of the engagement is a package: the DDL, the verification steps, the rollback and a fenced prompt, ready for your coding agent or for the person who will run it.
Why it matters: separating who designs from who executes is the only way an audit means anything. A tool that designs and applies on its own is checking its own work.
Project / Analytics layer / Deliverables
Execution package: create the analytics schema
Version 3 · for the client agent or the DBA
Environment assumptions
PostgreSQL 16 · user without SUPERUSER
run inside a transaction · no embedded credentials
if any check fails, apply the rollback and stop
Checks after applying
Options go into a matrix with weighted criteria and a justified score per criterion. And each one declares what almost nobody puts on paper: how much it costs and how long it takes to leave within 18 months, and exactly what locks you in.
Why it matters: being vendor-neutral only means something once you declare the exit. An option that does not declare shows up as "not declared", because a visible gap beats silence.
Where the analytics layer should live
4 weighted criteria · 40 GB active · team with no dedicated engineer
| Option | Score | Monthly cost | Cost to leave within 18 months |
|---|---|---|---|
| Analytical Postgres on the database you already run | 8.4 | USD 792 | Low Standard SQL, dump and restore. No proprietary format. |
| BigQuery + BI serverless | 7.9 | USD 468 | Medium Export to Parquet handles the data. Vendor-specific SQL and scheduled routines have to be rewritten. |
| Managed lakehouse full platform | 5.1 | USD 2,140 | High Catalog, permissions and notebooks have no equivalent elsewhere. Estimated 4 to 6 months. |
Recommendation: BigQuery + BI at this volume. Above 1 TB active, the comparison flips.
Every deliverable goes through an automatic audit before it is saved. The script runs on a disposable PostgreSQL, inside a transaction that is rolled back at the end. Decisions also face an adversarial reviewer, and a model from a different vendor gives the second opinion: where the two disagree gets published.
Why it matters: the confidence seal is not a probability of being right. It is the fraction of the available checks that was actually performed, with the weight and the reason for each one on screen.
Transactional model for e-commerce
Version 2 · audited · 6 entities, 31 columns
| DDL executed on a real database | done |
| Business questions answered by query | done |
| Adversarial review | done |
| Proof under load | not measured |
Second opinion, from a different model
Disagrees with normalising stock_movements: at high volume, a balance computed by summation becomes the bottleneck. Suggests materialising it. The disagreement was recorded, not resolved.
The cost projection does not end at the document. When the real invoice arrives, it is compared against what was projected, and the deviation feeds a figure visible inside the product: average deviation, share within the band and which way the error leans.
Why it matters: the project track record shows recommended against measured, including the cases where we were wrong. A field with no measurement shows a dash, never a zero: "nobody measured" is different from "measured and got zero".
Recommended against measured
Analytics layer · 4 months since rollout
| What was claimed | Projected | Measured | Deviation |
|---|---|---|---|
| Monthly cloud cost | USD 468 | USD 511 | +9.2% |
| Main dashboard query | < 2.0 s | 1.4 s | within |
| Daily ingestion | < 15 min | — | not measured |
| Assumption: volume below 1 TB | — | 61 GB | holds |
Cost accuracy on this project: within plus or minus 20% in 3 of 4 months, biased low.
The engagement
The engagement is not an open chat. It is a state machine with a visible ruler at the top of the screen, and moving from one phase to the next has a stated condition. The consultant proposes the change; you confirm it.
Asks until it understands the business, volume, latency, team and what already exists. Nothing is recommended here, on purpose.
Unlocks at 3 recorded requirements
Alternatives compared against weighted criteria, cost computed by tooling and assumptions recorded with the condition that invalidates them.
Unlocks when you confirm the direction
An assessment of the material you submitted, with findings by severity and the evidence inside the artefact itself. The only optional phase.
Only exists if you submitted an artefact
Data model with ERD and DDL, cost analysis, architecture document, execution prompt, proof-under-load plan.
Each one audited before it is saved
The conversation does not die. In come the report from whoever executed, the monthly invoice and the review of recorded assumptions.
Where the deviation becomes learning
Security and compliance
Every domain row is scoped by organisation in the database itself, with a test suite that proves isolation between tenants.
Passkey or TOTP, mandatory MFA if your organisation requires it, inactivity timeout and IP allowlist.
Roles from a versioned catalogue, and nobody grants what they do not hold. The last person with the admin role cannot be removed.
Mutations, security events and sensitive reads are recorded and queryable, visible to the administrator of your organisation.
EVODA never connects to your database to change anything. Reading your environment, when it happens at all, goes through the agent on your side, with a key scoped to a single project, and what comes back is a snapshot you sent.
How to start
The DDL, a structure dump or the export from your modelling tool. No real data: the structure is enough.
One minute
The points a senior review would hit first, each with the evidence inside your own schema and the practical impact.
About two minutes, no account
That is when the interview, the project and the memory begin. The diagnosis becomes the starting point instead of discarded work.
Entry by emailed link, no password
Frequently asked
Never to change anything. When the engagement needs a number from your environment, it produces an investigation script that you or your agent run, and the result comes back attached to the question that prompted it.
Because part of it is machine-verified and the rest is declared. The DDL runs on a real database, business questions become queries, and the confidence seal shows which checks were done and which were not.
No. No commission, no rebate, no partner programme. That is what lets the matrix recommend plain Postgres when the scenario calls for plain Postgres.
They stay in your organisation, isolated by construction, and are processed by third-party language models to produce the deliverables. The sub-processors are listed in the privacy policy.
Then their assistant will execute better than anything of ours, and we say so. EVODA is useful when the question is whether the platform is still the right one, what the current path costs, and what changes if the scenario changes.
The diagnosis is free and asks for no account. Plans are not published yet; talk to us and we will go through your case.
Paste the structure, get the most serious findings in about two minutes and decide afterwards whether an engagement is worth it. No account, no card.
Free diagnosis