NS
Nishant OS
Initializing Nishant AI
LOADING0%
All work
ProductionAI Banking Assistant · Multi-Agent · May 2026 – Jul 2026

NeuroBank

An AI banking assistant that runs real banking operations through conversation.

0
Specialized agents
NLU · Planner · Reflection · Execution
0
Core workflows
balance · transfer · history · advice
FastAPI
Backend
modular microservices
Secure
Auth
persistent state

NeuroBank is an AI-powered banking assistant built with Python, FastAPI, SQLite, and LLMs that automates banking services through natural-language interactions. A multi-agent architecture — NLU, Planner, Reflection, and Execution agents — handles intent recognition, reasoning, and secure task execution, with LLM function calling wired directly into backend APIs.

The problem

Everyday banking is buried in forms and menus. Natural language could automate it — but a single LLM call hallucinates, can't reason through multi-step tasks, and certainly can't be trusted to move money securely.

The solution

A four-agent pipeline decomposes each request: an NLU agent extracts intent, a Planner sequences the work, a Reflection agent verifies it, and an Execution agent invokes typed backend APIs via LLM function calling. A secure, modular microservice backend adds authentication and persistent state so operations stay reliable and extensible.

System Architecture
User Querynatural languageNLU AgentintentPlanner Agentsequences taskReflection AgentverifiesExecution Agentfunction callingBanking APIsFastAPI · SQLite
Challenges solved
  • 1Accurate intent recognition across varied natural-language phrasings.
  • 2Secure task execution — money movement must be verified, never guessed.
  • 3Reliable LLM function calling wired to real backend APIs.
Highlights
  • Four cooperating agents: NLU, Planner, Reflection, Execution.
  • LLM function calling bound to real backend banking APIs.
  • Secure, modular microservice backend with persistent state.
Technology
PythonFastAPISQLiteLLMsLangGraphFunction CallingMicroservices
Next project
ClarityStack
Local-First Knowledge Management