Flagship offering · Generative AI

Prompt BI: ask the warehouse, in plain English.

Shinrai’s vendor-agnostic GenAI business intelligence offering — the same governed natural-language experience on Snowflake (Cortex Analyst) and the AWS-native stack (Amazon QuickSight + Amazon Q in QuickSight).

Book a Prompt BI demo Talk to us
The pitch

Natural-language BI that actually works in production.

Every BI vendor in 2026 has a natural-language feature. Most of them stall before production for the same three reasons: the answers cannot be audited back to source, access controls do not respect row-level rules already in the warehouse, and the team is locked into one vendor’s GenAI surface.

Prompt BI is how we solve those three problems. It is not a tool we sell. It is the disciplined delivery of GenAI BI on the warehouse you already have — Snowflake, the AWS-native stack, or both — with the governance, lineage, and accuracy guard-rails that turn a feature into a production capability.

If your business needs an analyst, a credit officer, or a property manager to ask a real question and trust the answer, this is the offering.

Choose your warehouse

Same governed experience, your platform.

We deliver Prompt BI on whichever warehouse the business has standardised on — or both, side-by-side.

On Snowflake

Delivery surface: Snowflake Cortex Analyst against semantic models we define for your data domain.

Why pick this: the warehouse is already Snowflake; analysts work in Snowsight or BI tools that hit Snowflake natively; the governance and access controls already live there.

What we build: the semantic model, the lineage registration via Informatica CDGC, row-access policies and masking policies, accuracy guard-rails, and the user-experience layer.

On the AWS-native stack

Delivery surface: Amazon QuickSight with Amazon Q in QuickSight against trained Q topics over your AWS data lake or Redshift cluster.

Why pick this: the data already lives in S3 / Redshift / Athena; QuickSight is the standard surface for analysts and embedded use cases; IAM is the existing access-control system.

What we build: the Q topics, the row-level security and column-level security via QuickSight permissions, embedded-analytics integration where the use case is customer-facing, and the cost guard-rails.

What’s built in

Governance is not an afterthought.

Every Prompt BI delivery includes these from day one, on whichever warehouse you pick.

Lineage to source

Every answer traceable from the displayed number, through every transformation, back to the source system. Informatica CDGC where Informatica is in scope; Snowflake or AWS-native lineage otherwise.

Row-level access

The branch manager sees their branch. The regional head sees their region. The CFO sees the group. Enforced in the warehouse, respected by the GenAI surface — not bolted on at the prompt layer.

Accuracy guard-rails

The model is constrained to the semantic layer and Q topics we define. Out-of-scope questions return ‘cannot answer’ rather than a confident fabrication. Confidence and source-table citations visible on every answer.

Audit log of every question

Who asked, when, what the model returned, and what the user did with the answer. The same evidence trail a regulator expects for any other decision-support system.

Cost guard-rails

Per-user and per-team query budgets. Warehouse-credit ceilings before a runaway loop. The FinOps discipline GenAI workloads need.

Multi-warehouse where it matters

If the business has Snowflake and an AWS data lake, Prompt BI delivers the same experience on both. One semantic vocabulary, two delivery surfaces, no rewrite.

Where it lands

Industries we’re already deploying it in.

Financial services

A credit officer asks: “Which Nairobi-branch loans defaulted this quarter, by sector?” The answer arrives with source-table citations, the officer’s branch-level access enforced, and the full conversation logged for the audit trail. See the FS page.

Real estate

A property manager asks: “Which UAE units have rent due in the next 14 days?” Prompt BI returns the list grouped by building, with lease-document links via the EDMS layer. See the RE page.

Technology & SaaS

A customer-success lead asks: “Which accounts in the East-Africa region trended down on weekly active users this month?” The answer is delivered against the product-analytics warehouse, scoped to the CS team’s permissions. See the Tech page.

How we deliver it

Shape of a Prompt BI engagement.

1. Discovery (1–2 weeks). Which questions the business actually wants to ask. Which warehouse those answers live in. The deliverable is a question catalogue — not a tool selection.

2. Foundation (2–4 weeks). The semantic model (Snowflake) or Q topics (QuickSight). Lineage registration. Row-level access policies. Audit-log destination. The deliverable is a working environment.

3. First answers (4–6 weeks). The 10 most-important questions from Discovery, asked end-to-end, instrumented against the guard-rails. The deliverable is a working surface the business can use and audit.

4. Iteration (ongoing). More questions, more domains, more downstream consumers. Quarterly accuracy and cost reviews keep the model honest.

The shape is deliberately the same as our Trifecta data-platform delivery — Prompt BI is the natural-language layer on top of the platform.

Prompt BI by Shinrai Technologies
Shinrai Technologies — Flagship offering

Book a Prompt BI working session.

30 minutes with our team. Bring your three most-important business questions. We’ll show you the delivery shape on the warehouse you already have.

Snowflake Cortex Analyst delivery

Amazon QuickSight + Amazon Q delivery

Governance, lineage, and accuracy guard-rails from day one

Schedule a Consultation