Radency built an MVP of
a financial analytics SaaS with 7 AI agents
AI agents
SaaS
FinTech
5x
Lower token cost
7 AI agents
Plus an orchestrator
5 integrations
QuickBooks, Stripe, OpenAI API, Azure ML, Keycloak
Business context
We built an MVP of a SaaS financial analytics platform that connects to accounting systems, consolidates revenue and expense data, and provides AI-driven insights.
It answers questions on finances in plain language, visualizes results in dashboards, and forecasts future spending.
AI agents
that are able to predict company expenses?
Most companies still review expenses manually inside complex accounting tools.
With AI, this can be faster and more precise: agents can automatically pull data from accounting systems, explain it in plain language, and forecast future spend from recent patterns.
It sounds simple on paper. But building it isn’t. So we decided to start with an MVP.
“We wanted to pull data from our accounting systems and use AI agents to track and analyze the company’s revenue and expenses.”
Max Honcharuk
Partner & Solution Architect at Radency
Answer quality.
We had to turn long LLM outputs into concise, source-backed insights about the company's data.
Token cost.
We needed to control context size across agents to make analysis fast and affordable.

Data integration.
We had to sync QuickBooks and Stripe data and keep ledgers consistent.

Radency pulls off
a SaaS MVP powered by 7 AI agents
We started an R&D project to see how agentic AI performs on real financial data. In one month, we shipped the MVP, and a couple of weeks later, we polished it into a fintech SaaS with data-backed analysis.
1 →
We integrated QuickBooks & Stripe
We integrated accounting and payments first. QuickBooks and Stripe were the sources of truth for revenue and expenses, with reliable syncs to keep ledgers consistent.
2 →
First prototype with OpenAI done
Next, we prototyped an agentic loop: an insight agent pulled raw data, built context, queried the LLM multiple times, and returned a draft answer. We iterated until outputs were consistently correct.
3 →
We added ML-powered forecasting
Using the revenue/expense data, we trained a prediction model to project future spend. Every forecast shows where the numbers came from so finance teams can trust the output.
4 →
Split into 7 agents to cut token cost
Early runs sent too much context to the model, making answers slow and costly. We split work across 7 specialized agents, each processing only relevant context, cutting token use by ~5×.
5 →
Result: a working AI financial analysis tool
The app has role-based access with invites, token-based billing, AI alerts, and forecasts. 7 AI agents analyze financial data and surface spend spikes and areas for improvement.
Results achieved:
7 AI agents orchestrated
Specialized agents handle data prep, analysis, forecasting, and answer review in a single pipeline.
MVP in ~1.5 months
A small team built a finance analytics MVP with a dashboard, reports, and an AI chat in just several weeks.
Proven methodology
We proved that multi-agent AI can efficiently analyze financial data, explain it, provide insights, and forecast future spending.
“Now our team knows how AI agents behave on real-world data, so we can deliver agentic AI faster, more affordably, and with full confidence.”
Eduard Dolynskyi
Solution Architect at Radency
Finance analysis MVP, shipped in 1.5 months
We built a multi-agent finance app that ingests accounting and payment data, surfaces AI insights in a dashboard and reports, and lets users chat with their ledgers. It connects to QuickBooks and Stripe via secure APIs and runs on a cloud-ready microservices stack.
01
FEATURE
Agentic AI core
7 agents collaborate to prepare data, build context, call the LLM, and refine outputs into finance insights.
02
FEATURE
Dashboard with AI insights
The dashboard shows spend spikes, trends, and variances at a glance, with quick drill-downs to the underlying transactions.
03
FEATURE
Detailed reports & charts
Reports cover monthly, quarterly, and yearly views, and users can export them for reviews or sharing.
04
FEATURE
AI chatbot
Users can ask plain-language questions about P&L or expenses, and get answers grounded in the source data.
05
FEATURE
Explainable forecasting
The app projects future expenses from historical data, and each forecast shows where the numbers came from.
06
FEATURE
Role-based access & invites
Access is scoped by role, and team members can be added with simple invites.
Solution architecture:
.png)
Technologies used
→ AI / NLP
LLaMA
BGE Embedding Model
→ Backend
.NET 9
REST API (Flask)
→ Storage
PostgreSQL
Milvus
RabbitMQ
→ Integrations
QuickBooks
Stripe
LLamaIndex
LangChain
SendGrid
OpenAI API
Azure ML
Azure Blob Storage
Keycloak
From idea to agentic AI in finance in ~45 days
The platform is now powered by 7 AI agents that analyze, explain, and forecast a company’s revenues and expenses.
Before
No AI agents for finance analytics
High model cost from oversized context
Unclear projections and “black box” results
Intuition-based budget planning
After
7-agent MVP delivered in ~1.5 months
~5× lower token usage via smart routing
Explainable insights with cited sources
Data-driven forecasts of future spend