CRM for financial work · Human–AI collaboration

Turning experience in financial work into systems where people and AI work together.

gentry fixes the tangled workflows and information problems of existing customer-management systems. We work from workflow diagnosis and CRM design through building AI-assisted features, verification and roll-out.

Record, status, owner, schedule, documents and history connected on one screen. Rebuilt with sample data, not real customer information.

About gentry

A software company built on real financial-work experience

gentry is a technology company that builds customer-management and workflow-automation solutions based on hands-on experience in financial work.

We design work environments where consultation notes, customer history, progress status and schedules are connected. Drawing on our experience of rebuilding our own legacy admin into “NewAdmin”, we improve the screens and processes that staff use every day while preserving the data that has built up over years.

On top of that, we are adding AI that organizes information, drafts documents and assists with tasks, so that staff can review, decide and stay in control.

gentry started from the customer-management admin we used in loan-consultation work. The legacy system was hosted externally and could not be modified, and checking one customer's progress meant cross-checking status, signing and disbursement dates, lender results, recent activity and the next contact date separately.

Solutions and current status

What works today, and what we are building

Each item shows its actual status in NewAdmin. “In use” means it is used in real daily work; “In development” means being built or stabilized; “Planned” means not yet implemented.

CapabilityStatusWhat it does
Customer list and detail screensIn useStatus-based lists, saved custom lists and column settings, mobile customer cards and detail view.
Linked notes, schedule and statusIn useConsultation notes, signing and disbursement dates, next contact date and status in one customer record.
Data preservationIn useSource data is read-only; NewAdmin's own notes and statuses live in a separate store that sync never overwrites.
Role-based display of personal dataIn useThe server sends masked copies to non-admin accounts across lists, detail, search, data responses, Excel, print and mobile.
Lender result checksIn developmentResults are fetched through office browser sessions and saved to the record; session renewal is being stabilized.
Change review and full restore rehearsalIn developmentChanges are logged without personal values; a grouped change-review screen and restore rehearsal are part of the cut-over criteria.
Structuring free-text intakeIn useAn internal AI agent turns free-text intake messages into fixed fields; staff review a preview before anything is submitted.
Required-field checks, follow-up datesIn use RulesMissing required fields block the next step; recent records are checked by fixed rules to correct status and set the next contact date. No AI involved.
Consultation summaries, missing-item hints, draftsPlannedTo be built on the Claude API, shown as drafts and saved only after staff approval.

What AI does not do. It does not make loan underwriting, approval, rate or limit decisions, which belong to the lender. Submitting applications, changing a customer's status or contacting customers happens only after a staff member has checked the content.

Case study · Our own system · Started March 2026 · Running in parallel

Admin → NewAdmin: keeping years of customer history while redesigning how staff work

This is an improvement of the system we use ourselves, not a project for an external client and not a core-banking replacement. Switch between the old list and the new work board below.

Sample data · digit codes instead of personal data
  • Personal data shown to every account
  • Status, amounts and dates in separate columns
  • Lender results checked on other sites
  • No next action on screen
  • Progress at a glance
  • Next action and schedule
  • Automatic lender-result checks
  • Role-based masking

Background

  • The legacy PHP admin was externally hosted; its server and database could not be changed.
  • Progress information was scattered across status, dates, lender sites and notes.
  • Same-name customers, repeat applications and ten-year-old records all had to be kept.
  • Personal-data visibility had to differ by role, enforced on the server.

What we built

  • A new Python web application that reads the source data read-only and keeps the original IDs.
  • Its own store for notes, statuses and schedules, protected from sync overwrites.
  • Sync rules: source-owned fields follow the source, progress never moves backwards, valid missing dates are filled in.
  • Server-side masking for non-admin accounts, verified by automated tests.

Role of AI

  • Development: Claude Code and other AI coding tools for code analysis, writing failing tests, reviews, pre-deployment audits and documentation. Every change is backed up, tested and approved by a person.
  • Operations: an internal AI agent structures free-text intake; staff confirm before submission.
  • Product: the Claude API is not yet connected to NewAdmin. Planned features are listed above.

Where it stands

  • Built: lists, detail, status views, custom lists, mobile, role-based display, separated stores.
  • Verified: masking, preservation and sync-protection rules covered by automated tests.
  • Deployed: running on our server alongside the legacy admin, which remains the source system.
  • Not yet done: full cut-over criteria (7 consecutive days of successful sync, zero data drift, zero permission bypass, restore rehearsal, final sign-off).

We do not publish time-saving, revenue or error-rate figures for this case. We will add numbers only when they have been measured.

How we use Claude

Development tool, product feature, roadmap

Solid lines are connections running today in NewAdmin; dashed lines are planned Claude API connections.

  1. 01 Input Consultation notes, intake messages and lender results collect in the customer record. In use
  2. 04 · Staff Staff review A person checks and edits. Every AI suggestion must pass this step. In use
  3. 05 Approved action Only confirmed content leads to submission, status changes or customer contact. In use
  4. 06 Record the result Outcomes and changes are written back to the customer record. In use
Running today Planned (today, staff review directly after step 02)

Running today: paste the call notes, confirm in Telegram

Staff used to fill every box of an input form. Now they paste the notes they took, an internal AI agent sorts them into fields, and the staff member confirms in Telegram before anything is submitted.

Demo · digit codes instead of personal data
  • Every box typed by hand
  • Same data retyped for each lender
  • Results checked separately

1Paste the call notes

Called customer 000011222. Deposit 80M, no rent, balance due end of month so needs about 20M. Employed 3 yrs, credit average. Phone 000011222 ID 000011222 region 01 rental apt. Wants Lender A

As the staff member received it

2AI sorts it into 10 fields

  1. Name000011222
  2. ID000011222
  3. Phone000011222
  4. AddressRegion 01 rental apt
  5. DepositKRW 80M
  6. LoanKRW 20M
  7. LenderLender A
  8. PurposeBalance
  9. CreditAverage
  10. Notes3 yrs employed, no rent
In use · internal AI agent

3Staff confirm in Telegram

Admin intake preview
Please check the 10 fields.
1 Proceed2 Needs a fix3 Stop
1
✓ Admin intake done
Lender A preview checked → ✓ Submitted
Result: pre-approved KRW 20M
In use · runs only after staff confirm

4Saved to the NewAdmin record

Customer 000011222IntakePre-approved
  1. Intake
  2. Verify
  3. Pre-appr.
  4. Sign
  5. Disb.
  • Original text and structured fields saved
  • Lender result recorded · next: book signing
In use · automatic result checks being stabilized
  • Paste text, fields fill in
  • Confirm and submit from Telegram
  • Results linked to the record

Nothing is submitted until a staff member confirms the AI-structured fields in Telegram, and each lender submission gets its own preview first. The structuring is done by an internal AI agent; a Claude API integration inside NewAdmin is planned. This is a demo: names, ID numbers and phone numbers are replaced by digit codes.

Development tool · in use

Claude Code for code analysis, test writing, change review, audits, handover documents and this website, alongside other AI coding tools. People check test results before anything ships.

Product feature · not yet connected

NewAdmin and Bankly do not call the Claude API today. We built an internal, test-only MCP server that lets AI tools look up our loan-condition tables; it is not connected to any customer-facing service.

Roadmap · planned

Claude API for consultation summaries, missing-item hints and follow-up drafts. Only the fields a task needs are sent, never national ID numbers, and results are saved only after staff approval. We will test on sample data first, then run alongside the current process.

Company information and contact

gentry

Brand
gentry
Business entity
YJ Solution (Gentry Info) · 와이제이솔루션(젠트리인포)
Founder & CEO
Sangjun Jung
Founded
November 2024
Business Reg. No.
738-45-00483
Location
Gyeonggi-do, Republic of Korea
Email
taktwosj@gentryinfo.com

Tell us about your workflow

Send your company name, your name, a reply address and the work you want to improve. Please do not send real customer data; we scope projects from screens and workflows only.