Independent FP Office with Extensive Consultation Experience Develops and Releases RAG AI Agent 'Tally' from Scratch

Nagaoka FP Office LLC (Ginza, Chuo-ku, Tokyo; Representative Partner: Richi Nagaoka) has independently developed and released 'Tally,' an interactive AI agent for queries related to home purchases, mortgages, life insurance, and life planning.

The system also features a function to automatically generate household cash flow tables through dialogue alone.

All services can be used anonymously, for free, and without registration. Instead of having a general-purpose large language model answer directly, Tally employs a 'RAG (Retrieval-Augmented Generation)' structure. It generates responses by referencing an original knowledge source prepared by the company—a collection of documents systematizing 20 years of judgment criteria regarding home purchases, mortgages, life insurance, and life planning.

Based on the company's research, there are virtually no other FP offices or insurance agencies in Japan that have released a RAG-based AI agent incorporating their proprietary knowledge for consumers.

Introduction Page: https://gettally.jp/
Usage Page: https://nagaokafp.co.jp/tally-chat.html

[Key Points of This Release]

- Independent development and release of 'Tally,' an interactive AI agent specialized in housing, life insurance, and life planning.
- Adopts a RAG model reflecting the real-world consultation expertise of a veteran FP, rather than a general-purpose AI.
- Fully in-house development covering the front-end, API server, system instructions, and knowledge source.
- Implementation of 'automatic cash flow table generation via dialogue,' visualizing household finances up to age 95 on an annual basis.
- Available anonymously, for free, and without registration.
- Designed to improve response accuracy over time as the knowledge source is updated daily.
- Eliminates the psychological barrier of booking a consultation with a human FP, encouraging usage among younger demographics.

■ Independent Development and Release of RAG AI Agent 'Tally'

Nagaoka FP Office LLC has developed and released 'Tally,' an interactive AI agent that answers questions regarding housing, mortgages, life insurance, and life planning.

Users can receive professional answers comparable to those from a human FP 24 hours a day simply by typing questions into the chat screen.

■ Adopting a RAG Model Reflecting Real-World Consultation Expertise

Anyone can ask general-purpose AIs like ChatGPT about mortgages or insurance. However, the response is often a textbook-like average of general information found on the internet. Because general-purpose AIs are designed to always provide superficially neutral answers, they do not guarantee truly useful advice.

For example, a general AI might generate a shallow response like 'A safe mortgage borrowing amount is 5 to 6 times your annual income'—something you could learn from reading a web article. In reality, individuals want personalized answers regarding whether they can repay the loan under their specific circumstances and how much they are truly allowed to borrow.

Tally employs a 'RAG (Retrieval-Augmented Generation)' structure. When a user inputs a question, the AI reads 'system instructions' (hundreds of rules written by the company outlining response policies, covered domains, and prohibitions) while extracting relevant sections from the 'knowledge source' (a collection of judgment criteria formed through real-world consultations as a housing-specialized FP) via semantic search, and passes them to the response generation engine.

As a result, rather than providing an average AI response, it returns an answer reflecting 'how Nagaoka FP Office would respond to that question.'

Consequently, it can even generate an answer like, 'You should not take out a mortgage right now.'

Rather than acting as a general-purpose AI, it references the knowledge source every time to generate answers based on an FP's judgment criteria.

■ Fully In-House Development from Front-End to Knowledge Source

All code comprising Tally—from the front-end and API server to the system instructions dictating response policies and the knowledge source—was implemented in-house by the company.

'The task of verbalizing our experience and knowledge into a format that AI can read' is difficult to achieve through outsourcing. A significant amount of time was dedicated to preparing the knowledge source. Furthermore, automatic updates to the knowledge source have been incorporated as a feature.

■ Implementation of Automatic Cash Flow Table Generation via Dialogue

Tally automatically generates 'household cash flow tables' solely through dialogue—a task previously performed by human FPs using commercial FP software or life insurance company planning systems.

A cash flow table lines up a family's income, expenses, and savings on an annual basis, visualizing in numbers 'how much savings you will have at what age' and 'in which year you might fall into the red.'