01
Ingest
CSV, XLSX, JSON, Parquet. Streamed or batched. SHA-256 hashed on landing.
Foundation-model intelligence for spreadsheets and databases.
Five products on one engine — clean it, dedupe it, connect it, query it. Pick the ones you need; ship in an afternoon. Every output ships with an audit chain regulated industries can defend.
Built for sectors where the audit trail matters
HealthcareFinancePublic SectorPharmaInsuranceLegalOur Thesis · 2026
DAVA brings foundation-model intelligence to spreadsheets and databases — the lifeblood of science and business — and wraps every output in an audit chain regulated industries can defend.
Foundation models have transformed text and images; tabular data is the next surface, and the gating constraint isn't capability — it's trust.
That's what we're building.
The Engine
The same substrate, six stages, every job.
Every record passes through the same deterministic pipeline — ingest, profile, normalize, resolve, decide, evidence. Foundation-model intelligence where it earns its keep. Hash-chained lineage as a free byproduct.
02
Profile
Type inference, sensitivity tagging, outlier detection — the Smart Tables layer.
03
Normalize
Coerce types, fix encodings, fold case, parse dates. Sandboxed adapters.
04
Resolve
Cluster duplicates, surface FK candidates, suggest semantic links across files.
05
Decide
Assistive or Decisioning mode. Reviewer queue when human-in-the-loop.
06
Evidence
Signed ZIP pack: input hash, output, lineage, decisions, retention proof.
LLM Sandbox
Isolated
No customer data leaves the perimeter
Throughput
~3.4M rows/min
Per worker, on commodity hardware
Latency
p99 820 ms
Norm endpoint, 10k row payload
Modes
Two
Assistive · Decisioning
Five Products
Sold à la carte. Bought together when it matters.
Pick the products you need today. Add the rest later — they share auth, audit, and policies, so the second one is always cheaper to deploy than the first. No bundling pressure.
DAVA Norm
Drop a messy file, get a clean structured table. Type-inferred, encoding-fixed, sensitivity-tagged.
john Smith →John Smith
12/03/24→2024-03-12
€1,247.50→1247.50 EUR
DAVA Dedup
Cluster duplicate records across messy datasets. Fuzzy keys, canonical row selection, audit trail per cluster.
Read moreDAVA Connect
Discover relationships across datasets — FK candidates, value overlap, semantic links. The ring graph that finds the joins you forgot.
Read moreDAVA Agent
Talk to your data — chat with the dataset, or expose tools to your AI agents over MCP. Decisions logged like everything else.
Which leads converted in Q1?SELECT * FROM leads WHERE status='won'
AND quarter='Q1' → 142 rowsTop sources?
Read more AND quarter='Q1' → 142 rowsTop sources?
DAVA Trust
Lineage, audit chain, retention TTLs, evidence packs, EU AI Act mode. The compliance layer your CDO is going to ask about.
job.normalized · adapter=v3.214:22:01
policy.applied · pdp/redact-pii14:22:01
evidence.signed · sha256:a4f9…14:22:02
One substrate.
Same auth. Same audit chain. Same retention rules. Wire one product up — the next four cost a fraction of the integration time.
Shared services
→ auth · OIDC + SCIM→ audit · hash chain→ retention · per-org TTL→ policies · open PDP
Built In · Not Bolted On
Auditable by default. Not by add-on.
You're here for the AI products — clean, dedupe, connect, query. The audit chain is the part you didn't have to ask for: every transformation, decision, policy hit, and download writes a tamper-evident block to your org's chain. Replay any job, hand the evidence pack to your auditor — without ever leaving the platform.
01
Tamper-evident hash chain
Each event includes the previous event's hash. Break one and the chain visibly fractures.
02
Policy decision points
Open Policy Agent under the hood. Your PDP rules, your decisions, your records.
03
Evidence packs on demand
Signed ZIP — input hash, output, lineage graph, decisions, retention proof. Auditor-ready.
org.acme · live audit chain
12,847 events · last 24h
evidence.pack.signed
by
svc.evidence-builder for job norm.7f2asha256: a4f9c3…d802 ← prev: 8c1e0a…
14:22:08
just now
just now
policy.applied
rule
pdp/redact-pii · 412 columnssha256: 8c1e0a…b3f5
14:22:01
job.normalized
adapter
v3.2 · 2.41M rows · 3.2ssha256: 4d7b22…c91a
14:21:58
job.submitted
by
ana@acme.com · file customers_q4.csvsha256: 1e9ff0…7d34
14:21:54
key.minted
by
ana@acme.com · scope norm:writesha256: 9b2af1…e618
14:18:42
The Customer Portal
What your team actually uses.
Not a sales demo. The same dashboard your data engineers use every day — submitting jobs, reviewing clusters, minting keys, downloading evidence packs. Bring your own SSO, ship in an afternoon.
Welcome back, Ana
Jobs · Month
1,248+12%
Rows · Processed
42.7M
Sandbox · Today
8.4s
Audit · Events
12,847
Job
Mode
Rows
Status
customers_q4.csvjob.7f2a
Norm
2.41M
succeeded
leads.merge.csvjob.7f2b
Dedup
486K
running
orders & invoicesjob.7f29
Connect
12 files
succeeded
Pricing
Pay for the products you actually use.
No anonymous self-serve, no surprise overages. Every plan is contracted with AVA Research, with a one-week pilot included for new accounts.
Single Product
Ask AVA Research
Pick one product. Cloud-hosted in EU. Standard SLAs and support.
- One DAVA product, à la carte
- Trust layer included
- Email support · 24h SLA
- EU data residency
Bundle
Most pickedAsk AVA Research
All five products on the same engine. Volume scales with your data, not your line items.
- All five DAVA products
- Decisioning mode
- Priority support · 4h SLA
- SOC 2 Type II audit pack
- SSO · SCIM · custom roles
Self-host
Ask AVA Research
Docker bundle, JWT-licensed. Yours to deploy in your perimeter.
- Self-host · air-gap supported
- License JWT model
- Dedicated solutions engineer
- Custom retention policies
Bring us your hardest dataset.
We'll clean it, audit it, and give you back a file you can ship — plus an evidence pack you can hand to your auditor. Most pilots run in under a week.
no credit card · no anonymous signup · contracts with AVA Research SRL