The Platform · Launching soon
Titan by Tax.AI
Titan is the platform underneath Tax.AI — the AI-native compliance and advisory engine room. It turns GCC tax law into deterministic engines, checks its AI against the engines’ figures, and writes service runs to an append-only audit log, Merkle-sealed with public inclusion proofs.
§ 01How a position is produced
Ingest → Compute → Advise → Seal → Defend
Ingest
Financials and documents in; entity profile and facts established.
Advise
Grounded AI narrates the verified result — planning levers quantified by re-running the engine.
Seal
Service runs are written to the append-only audit log, sealed into Merkle batches with public inclusion proofs.
Defend
Audit-defence pack on demand: the audit log, trace hashes and inclusion proofs, with a file manifest a reviewer can check offline.
§ 02Schedule of modules
Six modules, one safety contract.
1.1
Verified engine suite
UAE CT, KSA Zakat/WHT and Pillar Two/DMTT as tested, deterministic engines, plus transfer-pricing documentation thresholds and an indicative margin screen.
1.2
Grounded advisory core
One safety contract across every AI memo: its figures are checked against the engine’s result, and on a mismatch the computed result is returned instead.
1.3
Tax Copilot
Conversational engine tool-use: the copilot computes with a verified engine instead of letting the model guess the number.
1.4
Forensic provenance
Append-only audit log, Merkle-sealed in batches, with a public inclusion-proof verifier.
1.5
Compliance operations
E-invoicing (UAE UBL 2.1 validation and XML; KSA ZATCA QR encoding and FATOORA rule checks), CbCR data preparation and XML draft, compliance calendar, penalty projection, audit-defence export.
1.6
Tenant-isolated by design
Row-level security on tenant tables, and tenant scoping enforced in application code, guarded by lint and tests, on the server paths that bypass it. Benchmarking reads only the peer figures you supply.
§ 03Grounded AI
Computation before narration.
Grounded AI advisory
The AI writes over a verified engine result. A grounding gate checks the memo’s figures against that result and your stated facts; on any mismatch the memo is discarded and the deterministic result is returned.
Tax Copilot with live engine tool-use
Ask in plain language. The copilot routes the question to a verified engine, runs it and answers from the engine’s figures. If its prose cites a number outside the result, the prose is dropped and the computed result is shown.
Law-change modelling
Change the UAE Corporate Tax rate or the 0% band and re-run the verified engine across up to 200 entities, with each change labelled by whether it cites a verified legal source.
Peer benchmarking
Place your effective tax rate or readiness score within a set of peer figures you supply: median, quartiles and your percentile, with every statistic withheld below five peers.
§ 04Defensible by construction
The record behind the result.
Deterministic engines
Statutory computations run in tested code, not prompts. The same inputs, including the as-of date, give the same figures, with the legal basis attached.
Forensic Merkle seal
Audit-log entries are sealed into Merkle batches by a scheduled job. Anyone with an entry’s hash can check its inclusion on our public, no-login verifier.
Tier-labelled authorities
Legal bases in advisory memos and client deliverables carry an evidence tier (verified primary law, official guidance or unverified) and link to the official source when it is in our verified register.
Audit-defence packs
Export an evidence pack: the audit log, AI-call trace hashes and per-entry inclusion proofs, with a SHA-256 file manifest a reviewer can check offline.
Ask Titan a tax question. Watch it run the engine instead of guessing.
Titan is in its final build and opens to early seats first. Book one and we will contact you before it opens.