From Noise to Insight: How LLMs Are Transforming Threat Intelligence Analysis

Flipkart's security team built DeepInsight, an LLM-powered platform to automate threat intelligence analysis across 200+ daily alerts. The system ingests unstructured data from multiple sources, normalizes it, and uses Gemini (chosen over GPT-4o and an internal model for its large context window and pricing) to generate structured summaries with true/false positive indicators and recommended actions. Prompt refinement was critical, improving critical data extraction from 60% to 85%. The platform reduced manual triage time by ~75%, enabling analysts to focus on investigation rather than formatting. Future plans include benchmarking Claude and newer Gemini models, and expanding to additional alert types.