aegisheimdall/AEGIS-FIN-1

TEXT GENERATIONConcurrency Cost:1Model Size:7BQuant:FP8Ctx Length:4kPublished:Mar 8, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

AEGIS-FIN-1 is a Mistral-7B-Instruct-v0.3 model fine-tuned by aegisheimdall using QLoRA for specialized financial AI tasks. This 7 billion parameter model excels at structured transaction reasoning, including financial intent classification, bill negotiation, and credit optimization. It is designed to operate as an intelligence layer within a multi-stage safety pipeline, generating structured JSON responses for downstream processing. The model was trained on approximately 100,000 synthetic financial queries, focusing on safe and compliant financial assistance.

Loading preview...

What is AEGIS-FIN-1?

AEGIS-FIN-1 is a specialized financial AI assistant developed by aegisheimdall, built upon the mistralai/Mistral-7B-Instruct-v0.3 base model. It has been fine-tuned using QLoRA (4-bit NF4 quantization) on approximately 100,000 synthetic financial queries to provide safe and structured reasoning for various financial tasks.

Key Capabilities

  • Financial Intent Classification: Accurately identifies user intents related to purchases, bill negotiation, credit optimization, and Buy Now Pay Later (BNPL).
  • Structured Output: Generates structured JSON responses, making it suitable for integration into automated financial pipelines.
  • Safety-Oriented: Designed to operate within a multi-stage safety pipeline that includes content classification, injection detection, PII detection, and compliance validation.
  • Domain-Specific Knowledge: Covers topics such as credit cards, BNPL, budgeting, credit building, gig worker finance, and basic tax information.

What makes THIS different from other models?

Unlike general-purpose LLMs, AEGIS-FIN-1 is purpose-built for the financial domain with a strong emphasis on safety and structured output. Its fine-tuning on synthetic financial queries, calibrated against U.S. Federal Reserve data, allows it to understand and process complex financial scenarios with high accuracy (94.5% intent classification accuracy). The model's integration into a robust safety pipeline is a core differentiator, ensuring that financial advice is compliant and secure, actively blocking PII and potential fraud attempts.

Should I use this for my use case?

AEGIS-FIN-1 is ideal for applications requiring a specialized AI for financial transaction classification, negotiation, and optimization within a controlled, safety-first environment. It is particularly suited for:

  • Financial AI assistants: Providing structured recommendations for personal finance management.
  • Automated financial processing: Classifying and routing financial queries for backend systems.
  • Compliance-focused applications: Where input validation, PII detection, and audit trails are critical.

However, it is not intended for:

  • General-purpose conversations or chatbots.
  • Medical, legal, or non-financial advice.
  • Direct execution of financial transactions (it only recommends).
  • Use outside of a safety pipeline, as it assumes upstream filtering and validation.

Limitations to consider:

  • US-centric: Training data is calibrated for the U.S. market, limiting applicability to other regulatory jurisdictions.
  • Synthetic data: May have limitations in covering all real-world edge cases.
  • LoRA adapter only: Requires the base Mistral-7B-Instruct-v0.3 model for inference.

Finetunes
1 models