Build AI Credit Analyst Bot
Budget: $250 – $750 AUD
I want to spin up a data-analysis bot that can be trained to act as a credit analyst. Its core duties will be credit risk assessment, financial reporting, and running predictive analysis on lending portfolios.
I have not yet settled on the underlying LLM—Claude, GPT-4, or possibly a hybrid approach—so part of this engagement is your guidance on the best stack (e.g., LangChain orchestration, vector store, retrieval strategy) for accuracy, cost-efficiency, and explainability.
Here is what I need to walk away with:
• A working agent hosted in a cloud environment (or locally if more practical) that ingests structured and unstructured financial data, runs the three tasks above, and produces clear, exportable outputs.
• A concise technical document covering architecture, model choices, prompt chains, and instructions for retraining or fine-tuning with new data.
• A brief demo session so I can see the workflow end-to-end and ask questions.
Success looks like:
The AI agent can ingest a real-world loan scenario, assess it like a conservative private credit analyst, and output a structured credit memo aligned with our standards.
It must:
• Identify key risks (blunt and commercial, not generic)
• Flag high-risk borrowers/guarantors using data + external checks (e.g. adverse history, inconsistencies)
• Apply the 5 C’s of Credit with clear rationale
• Assess LVR, security strength, and exit strategy viability
• Recommend pricing (rate + fees) based on risk
• Produce a clean, investment-grade credit report (PDF/CSV)
• Generate forward-looking default risk indicators
The output should mirror how a senior credit analyst in a non-bank lender would actually approve or decline deals — focused on downside protection, speed, and clarity.
I have not yet settled on the underlying LLM—Claude, GPT-4, or possibly a hybrid approach—so part of this engagement is your guidance on the best stack (e.g., LangChain orchestration, vector store, retrieval strategy) for accuracy, cost-efficiency, and explainability.
Here is what I need to walk away with:
• A working agent hosted in a cloud environment (or locally if more practical) that ingests structured and unstructured financial data, runs the three tasks above, and produces clear, exportable outputs.
• A concise technical document covering architecture, model choices, prompt chains, and instructions for retraining or fine-tuning with new data.
• A brief demo session so I can see the workflow end-to-end and ask questions.
Success looks like:
The AI agent can ingest a real-world loan scenario, assess it like a conservative private credit analyst, and output a structured credit memo aligned with our standards.
It must:
• Identify key risks (blunt and commercial, not generic)
• Flag high-risk borrowers/guarantors using data + external checks (e.g. adverse history, inconsistencies)
• Apply the 5 C’s of Credit with clear rationale
• Assess LVR, security strength, and exit strategy viability
• Recommend pricing (rate + fees) based on risk
• Produce a clean, investment-grade credit report (PDF/CSV)
• Generate forward-looking default risk indicators
The output should mirror how a senior credit analyst in a non-bank lender would actually approve or decline deals — focused on downside protection, speed, and clarity.