AI-native fraud defense for modern analyst teams, combining anomaly scoring, behavioral identity, and network intelligence in one unified workspace.

Product Preview · Getting Started · Technical Documentation · Issues

Current preview: a full-screen network replay view for tracing suspicious money movement and exposed mule paths.

Sentinel is an AI-assisted fraud analyst workspace for detecting suspicious transfers, triaging high-risk alerts, and investigating connected money movement in real time. It is built to show not just that a transaction is risky, but why it was flagged, what pushed it from review to hold or block, and how network exposure changes the investigation.

The product combines a synthetic live stream for demos, a score-first incident queue, explainable decision logic, graph-based investigation, business-impact framing, and CSV upload analysis for ad hoc datasets. The result is a more presentation-ready fraud console that still reads like a real professional tool.

• Traditional fraud tooling often scores isolated transactions well, but makes it harder to see how signals interact across behavior, device, and network context.
• Analysts need a single workspace for triage, explanation, and investigation instead of switching across separate dashboards and graph tools.
• Demo environments need reliable, repeatable fraud scenarios that still feel realistic enough to tell a strong product story.
• Sentinel addresses those gaps with anomaly scoring, deterministic rules, and network analysis inside one unified analyst workspace.

1. Detect anomalous transfers and suspicious clusters in a live synthetic stream.
2. Prioritize incidents in a queue built for triage, filtering, and scenario injection.
3. Open the triage dock to review the combined transaction, behavior, and network score breakdown.
4. Move into investigation views to trace money flow, exposed entities, and suspicious graph patterns.
5. Support the narrative with business-impact metrics, upload-based analysis, and optional AI explanations.

• Next.js 15
• React 19
• TypeScript
• Tailwind CSS
• Cytoscape.js
• Recharts
• React Three Fiber

• FastAPI
• Python
• pandas
• networkx
• python-dotenv
• python-multipart

• OpenAI-compatible chat and explanation flows
• Deterministic fallback behavior when no live model is configured
• Incident and transaction chat surfaces grounded in current case data

• Python 3
• Node.js and npm

git clone https://github.com/SarveshwarSenthilKumar/GenAI-Genesis.git
cd GenAI-Genesis

cd backend
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
uvicorn app.main:app --reload

The backend runs at http://127.0.0.1:8000.

cd frontend
npm install
npm run dev

The frontend runs at http://127.0.0.1:3000.

The backend reads from either the repo root .env file or backend/.env.

Backend example:

FRONTEND_ORIGINS=http://localhost:3000,http://127.0.0.1:3000,http://localhost:3001,http://127.0.0.1:3001
OPENAI_API_KEY=
OPENAI_MODEL=gpt-4o-mini
OPENAI_TIMEOUT_SECONDS=8
OPENAI_BASE_URL=https://qyt7893blb71b5d3.us-east-2.aws.endpoints.huggingface.cloud/v1

Frontend example:

NEXT_PUBLIC_API_BASE_URL=http://127.0.0.1:8000

If NEXT_PUBLIC_API_BASE_URL is unset, the frontend defaults to http://127.0.0.1:8000.

• npm run dev starts the development server
• npm run dev:fresh clears .next and starts fresh
• npm run reset clears .next
• npm run build creates a production build
• npm run start serves the production build

Sentinel is strongest when it is shown as a complete investigation loop rather than a single score. The backend layers several systems together:

• Transaction scoring: flags unusual payment amounts, velocity, and transfer behavior.
• Behavior scoring: captures session drift, device changes, and baseline deviations.
• Network scoring: detects exposure to suspicious clusters, rapid chains, fan-in/fan-out, and cash-out paths.
• Decision transparency: exposes the top driver, tipping point, counterfactual context, and analyst-friendly explanation.
• AI assistance: enhances incident and transaction review, while deterministic flows keep the core demo reliable without a live model.

Selected API surfaces that power the workspace:

• GET /api/incidents/queue
• GET /api/incidents/{incident_id}/panel
• GET /api/incidents/{incident_id}/graph
• GET /api/live/bootstrap
• GET /api/live/stream
• GET /api/live/scenario
• POST /api/uploads/transactions/live
• POST /api/uploads/transactions/dashboard

• Expand the scenario library for more repeatable fraud-story demos.
• Continue refining the triage and investigation experience around explanation quality and analyst actions.
• Deepen upload-based reporting so ad hoc datasets feel as polished as the live monitor.
• Keep iterating on graph exploration, including the experimental 3D network surface.

• Repository: SarveshwarSenthilKumar/GenAI-Genesis
• Project summary: PROJECT_INFO.md
• Technical deep dive: COMPREHENSIVE_DOCUMENTATION.md
• Issues and enhancements: GitHub Issues
• License status: no license file is currently included in this repository.

(back to top)