AI Fraud Detection Platform for Banking
Budget: $1,500 – $3,000 USD
Overview:
We are building an AI-based platform that will automate IT audit tasks inside banks, insurance companies, and large enterprises.
The core objective is to replace or accelerate manual IT audit work by using local LLMs (Large Language Models) that analyze logs, reports, access controls, and system events – and generate technical comments, risk assessments, and audit findings.
IMPORTANT:
This tool must run fully on-premise, without any connection to the cloud. Data privacy and compliance (GDPR, DORA, ISO 27001, SAE 3402) are critical. The model must be embedded locally and work offline.
Core Goals
• Allow banks to upload CSV / Excel / Log / PDF files from internal systems (SAP, Active Directory, Database logs, SIEM, etc.)
• Automatically analyze access patterns, permissions, user behavior
• Use a local LLM (like Mistral, LLaMA, GPT4All) to generate:
• Audit commentary (e.g. “User X accessed SE38 outside of business hours – risk HIGH”)
• Risk levels (Low/Medium/High)
• Recommendations and findings for reports
• Output must be saved as PDF / DOCX / XLSX
Technology Stack (Proposed)
• Backend: Python, Docker
• LLM: Mistral 7B, LLaMA, Phi-3 (via Ollama or Hugging Face Transformers)
• AI Framework: LangChain, LlamaIndex, or similar
• Frontend (MVP): Streamlit or Flask (clean UI)
• Output: python-docx, openpyxl, fpdf
• Deployment: Fully on-premise via Docker or VM
We are building an AI-based platform that will automate IT audit tasks inside banks, insurance companies, and large enterprises.
The core objective is to replace or accelerate manual IT audit work by using local LLMs (Large Language Models) that analyze logs, reports, access controls, and system events – and generate technical comments, risk assessments, and audit findings.
IMPORTANT:
This tool must run fully on-premise, without any connection to the cloud. Data privacy and compliance (GDPR, DORA, ISO 27001, SAE 3402) are critical. The model must be embedded locally and work offline.
Core Goals
• Allow banks to upload CSV / Excel / Log / PDF files from internal systems (SAP, Active Directory, Database logs, SIEM, etc.)
• Automatically analyze access patterns, permissions, user behavior
• Use a local LLM (like Mistral, LLaMA, GPT4All) to generate:
• Audit commentary (e.g. “User X accessed SE38 outside of business hours – risk HIGH”)
• Risk levels (Low/Medium/High)
• Recommendations and findings for reports
• Output must be saved as PDF / DOCX / XLSX
Technology Stack (Proposed)
• Backend: Python, Docker
• LLM: Mistral 7B, LLaMA, Phi-3 (via Ollama or Hugging Face Transformers)
• AI Framework: LangChain, LlamaIndex, or similar
• Frontend (MVP): Streamlit or Flask (clean UI)
• Output: python-docx, openpyxl, fpdf
• Deployment: Fully on-premise via Docker or VM