GenerativeX

Financial Services / Mizuho Financial Group

Proposal Generation Engine for Mizuho RM Studio

presence.ai was selected as the proposal generation engine inside Mizuho RM Studio, Mizuho Financial Group's AI application suite for corporate relationship managers.

Mizuho Financial Group selected presence.ai as the proposal generation engine inside Mizuho RM Studio, its integrated AI application suite for corporate relationship managers.

Initial rollout to approximately 800 RMs began in February 2026, with planned coverage extending to all 3,500 corporate RMs firmwide.

The system turns structured inputs into presentation-ready PowerPoint slides that follow Mizuho's templates, color palette, and charting standards.

800 RMs

Initial rollout

Rollout began in February 2026

3,500 RMs

Planned coverage

Corporate RMs firmwide

50%

Prep-time reduction target

Target set for RM Studio

3x / 2x

Commercial activity targets

Client touchpoints and visit volume

The Operating Problem

Corporate relationship managers needed faster preparation without losing proposal quality.

Corporate bankers face a time-intensive preparation workflow before every client meeting, from company and industry research to hypothesis development, slide creation, post-meeting notes, and CRM updates.

Mizuho RM Studio was built as an integrated AI application suite to support that end-to-end RM workflow. The proposal generation layer is critical because it turns upstream intelligence into client-ready material.

What Was Built

presence.ai powers the last-mile proposal generation layer inside Mizuho RM Studio.

Upstream RM Studio agents gather and structure data such as client financials, sector trends, and Mizuho's product catalog.

presence.ai transforms that structured input into presentation-ready slides that comply with Mizuho's brand templates, color palette, and charting standards. Charts are rendered as native PowerPoint objects so RMs can fine-tune figures or styling before client use.

How It Shipped

The work was embedded in a production enterprise AI platform, not left as a standalone demo.

Mizuho RM Studio is an internally developed AI application suite for corporate RMs. GenerativeX's presence.ai was selected as the proposal generation engine within that environment.

The deployment shows the value of a focused last-mile generation layer: upstream agents can produce structured intelligence, while RMs receive editable, brand-compliant outputs that support real client conversations.

Scale Path

The rollout is tied to measurable enterprise adoption targets.

Initial rollout to approximately 800 RMs began in February 2026, with plans to extend coverage to all 3,500 corporate RMs firmwide.

Mizuho FG has set targets for RM Studio to cut pre-meeting preparation time by 50%, triple client touchpoints, and double visit volume.

Life Sciences / Life sciences workflow

Medical Review

An AI-assisted review workflow that compares machine output with manual review and exports structured results.

The application accepts medical PDFs for AI review and manually reviewed PDFs for OCR extraction.

Users can run single or bulk AI/manual review jobs, compare outputs, edit categorized differences, and export to Excel.

The architecture uses a Next.js frontend, FastAPI/Celery backend, and OpenAI Agents SDK for configurable multi-agent review execution.

AI + manual

Review outputs compared

AI findings validated against manual review results

Bulk jobs

Execution model

Single and batch review workflows

Excel

Structured export

Categorized differences exported for downstream use

Next.js / FastAPI

Technical foundation

With Celery and OpenAI Agents SDK

The Operating Problem

Review quality depends on seeing where AI and human reviewers differ.

The workflow starts with two controlled inputs: medical documents for AI review and manually reviewed PDFs for OCR extraction.

The design supports review quality by validating AI findings against manual review results, while improving efficiency through dashboards, batch processing, and structured exports.

What Was Built

A single workflow connects upload, review execution, comparison, and export.

Users can launch manual review, AI review, and comparison from one page with reusable execution blocks.

The comparison workflow surfaces categorized differences, lets users edit them, and exports results to Excel.

Agent Controls

The system keeps agent behavior configurable for regulated review work.

Teams can configure manager and worker instructions, agent patterns, model choices, and search options.

The technical foundation combines a Next.js frontend, FastAPI/Celery backend, and OpenAI Agents SDK.

Impact

The value is operational control, not just faster generation.

The workflow targets two impact areas: review quality through AI/manual validation and efficiency through dashboards, batch processing, and structured exports.

The workflow is designed around review, comparison, and export steps that keep humans in control of the final work product.

Start with one workflow. We'll show what can be built, governed, and adopted.

See what's possible