Skip to content

Noloopai/noloop-docs

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

9 Commits
 
 
 
 
 
 
 
 

Repository files navigation

NoLoop

One shared record. One fast decision. No loop.

NoLoop is a multi-agent AI platform for health-insurance claim processing. It sits in the middle of the ecosystem — patient, hospital, TPA, and insurer — on one shared system, so a claim never has to bounce back and forth between three parties again.

Each stakeholder gets their own website, and the patient is served through a WhatsApp bot (no app to install). A minimal patient web dashboard comes later.


The problem — "The Loop"

When a patient is admitted and wants to use health insurance, the claim travels through multiple hands:

Patient admitted → Hospital → TPA → Insurer → query back → Hospital → Insurer → query → …

This back-and-forth is The Loop. Three things make it painful:

  1. Documents are requested again and again.
  2. No shared status — three parties keep three separate files and three different pictures of the same claim.
  3. The patient pays cash to leave — "cashless is pending, pay now, we'll refund later." Later becomes weeks, months, sometimes never.

Meanwhile the patient occupies a bed another emergency patient needs.

The scale (India)

  • 11–12.5% of claims are denied, usually without a clear reason to the patient.
  • 58%+ of claims are cashless (hospital paid directly) — the dominant, growing method.
  • A clean cashless claim takes 2–3 hours; a problematic one takes days.
  • One human processor handles 50–80 claims/day, by hand.
  • ~15% of claims involve fraud → ₹8,000–10,000 cr in annual losses.
  • Dec 2025: IRDAI penalised an insurer ₹1 cr for processing delays and poor transparency. The regulator is watching.

The solution

NoLoop is one shared platform — think a shared document every stakeholder can see — that works in three steps:

Step What happens
1. Capture Hospitals submit structured claims. NoLoop catches missing info before submission, so claims aren't bounced back for errors.
2. Decide Policy rules (coverage, limits, exclusions) are auto-checked. High-confidence claims are decided instantly; the rest go to a human reviewer with an AI summary.
3. Settle The insurer pays the hospital directly (UPI). The patient pays only their co-pay and walks out.

Products — a separate site per stakeholder

Each stakeholder gets a focused product, deployed to its own subdomain. All sites share one component library and talk to the same backend and shared claim record.

Product Who What it does Status
Admin site Platform ops Manage orgs & users, credentials, audit logs (god-mode) Built (scaffold)
Hospital site Billing desk Submit claims/pre-auth, track status, manage beds/patients Scaffold → MVP
Insurer site Reviewing doctors + TPA role Review queue, approve/override, TAT dashboards Scaffold → MVP
Patient — WhatsApp bot Policyholders Real-time status updates + policy Q&A, no app to install Planned
Patient dashboard Policyholders Minimal web view of the claim timeline Later
TPA site TPA teams Dedicated TPA workspace (only if the workflow needs it) Later / optional
Landing site Public Marketing + sign-up Built

Why not a separate TPA site from day one? For a small team shipping fast, TPA is a role inside the insurer site to start with. It graduates to its own site only if the workflow genuinely demands separation.

Three stakeholders, three benefits

  • Patient — real-time WhatsApp updates at every step; no app to install; knows the exact co-pay before discharge.
  • Hospital — one console for all insurers; file once, watch it clear; faster bed turnover.
  • Insurer / TPA — clean structured claims instead of messy PDFs; AI-assisted review (target 3× reviewer throughput); every decision recorded and auditable.

The four AI agents

NoLoop's intelligence is organised as four cooperating agents. AI assists; a human makes every final approval decision (human-in-the-loop). (All four are built within the 8-week delivery plan — see the roadmap.)

Agent Serves Job
Query-Proofing Agent Hospital Cross-checks documents against the policy before submission and flags mismatches instantly.
Case Review (Adjudication) Agent Insurer Reads medical files and summarises the case in plain English for the reviewing doctor.
Fraud Detection Agent Insurer Turns billing patterns and history into an explainable fraud-risk score.
Communication Agent Patient Powers the WhatsApp bot's updates and policy Q&A, citing exact policy clauses.

Architecture (overview)

   Clients:  Hospital site · Insurer site (+TPA) · Admin site · Patient WhatsApp bot
                                   │
                    ┌──────────────▼───────────────┐
                    │  Core Services (API Gateway)  │  auth/RBAC · Policy · Claim · Notification
                    └──────────────┬───────────────┘
                    ┌──────────────▼───────────────┐
                    │     Workflow Orchestrator     │  state machine over the claim lifecycle
                    └──────────────┬───────────────┘
                    ┌──────────────▼───────────────┐
                    │      Shared Claim Record      │  single source of truth
                    └──────────────┬───────────────┘
        ┌──────────────────────────┼──────────────────────────┐
        ▼                          ▼                          ▼
  AI & Rules Engine        Data Layer                  Integrations
  4 agents + RAG +         Postgres · Docs ·           UPI settlement ·
  Rules Engine             Redis · Audit · Analytics   WhatsApp Business API

Full layered design, claim state machine, and event flow: ARCHITECTURE.md.


Tech stack

Layer Technology
Frontends Next.js / React, TypeScript, Tailwind CSS — one shared UI library, separate sites per stakeholder, white-labelled per client (e.g. apollo.noloop.in)
Backend API FastAPI (Python) — agent workflows; NestJS + Prisma service (current)
AI layer Groq LLM via API · RAG over 3 knowledge bases (policy docs, clinical billing benchmarks, insurance rules)
Documents OCR + AI extraction (handles handwritten prescriptions)
Database / cache PostgreSQL (source of truth) · Redis (cache)
Messaging Event bus / message broker · Webhooks
Patient comms WhatsApp Business API (bot)
Settlement UPI payment rails
Infra Docker · CI/CD (GitHub Actions) · cloud deployment

Monorepo layout

Today the hospital, insurer, and patient views live as routes inside noloop-app; the admin site is already separate. The roadmap splits them into their own deployable sites (target layout shown with ).

noloop/
├── noloop-app/                 # Today: landing + hospital/insurer/patient routes + services
│   ├── app/                    #   → splits into noloop-hospital, noloop-insurer, patient dashboard
│   ├── components/ · lib/      #   → shared UI library used by all sites
│   ├── backend/                # NestJS + Prisma API (current primary)
│   │   ├── prisma/             #   schema.prisma + migrations
│   │   ├── scripts/            #   create-platform-admin, seed-demo, generate-synthetic-claims
│   │   └── src/                #   auth, admin, claims, beds, metrics, catalog, ai, health
│   ├── backend-py/             # FastAPI API (target backend for agent workflows)
│   └── ai/                     # FastAPI AI engine (pipeline: extract→coverage→validate→adjudicate→llm)
├── noloop-admin/               # Admin site (Next.js): orgs, users, logs
└── (planned) whatsapp-bot/     # Patient WhatsApp bot service

Services & ports

Service Path Port Start command Status
Landing / web noloop-app 3000 bun run dev Built
Admin site noloop-admin 3001 bun run dev Built
Hospital site noloop-hospital (planned) 3002 bun run dev Planned
Insurer site (+TPA) noloop-insurer (planned) 3003 bun run dev Planned
WhatsApp bot whatsapp-bot (planned) 3005 bun run dev Planned
Backend API noloop-app/backend 4000 bun run dev Built
Backend (Py) noloop-app/backend-py 8001 uvicorn app.main:app --reload Built
AI engine noloop-app/ai 8000 ./start.sh Scaffold

Getting started

Prerequisites

Node.js 20+ and Bun (or npm) · Python 3.11+ · a Supabase / PostgreSQL database.

1. Backend API + database

cd noloop-app/backend
bun install
cp .env.example .env            # set DATABASE_URL + JWT secret
bunx prisma migrate deploy      # apply migrations
bun run scripts/create-platform-admin.ts   # bootstrap the first admin
bun run scripts/seed-demo.ts               # seed demo orgs, policies, patients, beds, claims
bun run dev                     # → http://localhost:4000

2. AI engine

cd noloop-app/ai
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env            # set GROQ_API_KEY (optional; runs without it)
./start.sh                      # → http://localhost:8000

3. Frontends

cd noloop-app   && bun install && bun run dev     # → http://localhost:3000
cd noloop-admin && bun install && bun run dev     # → http://localhost:3001

Quick reference: noloop-app/start.md.


Feature status

Built

Landing page · sign-up & login · JWT auth · RBAC · multi-tenant architecture · custom staff email generation · organization & employee management · admin dashboard, analytics & audit trail · hospital & insurer portal scaffolds · backend API (NestJS + Prisma) + FastAPI parity backend · Supabase integration · synthetic claim-data generator · AI pipeline scaffold.

Planned

Separate hospital / insurer sites · WhatsApp patient bot · workflow orchestrator (state machine) · single shared claim record · rules engine · the 4 AI agents · OCR + RAG · confidence scoring · UPI settlement · patient dashboard · TPA site (optional) · Redis · event bus · webhooks · full test pyramid, security, CI/CD, monitoring, deployment.

See ROADMAP.md for the full 8-week (2-month) delivery plan.


Documentation

  • ROADMAP.md — the full 8-week (2-month) Agile plan for a 4-person learning team: every feature and its tests, sprint by sprint.
  • ARCHITECTURE.md — layered system design, state machine, event flow.

How the vision maps to reality

This README describes the full product vision. The foundation — auth, multi-tenancy, portals, and API/AI scaffolding — is built today; the separate sites, WhatsApp bot, multi-agent intelligence, orchestration, settlement, and full test suite are delivered across the 8-week roadmap.

Human-in-the-loop: NoLoop's AI never issues a final approval on its own — it reads, flags, summarises, and scores; a human reviewer signs off. Product claims such as "92% decided instantly" and "pre-auth in 8 minutes" are target outcomes, not measured production metrics.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors