SecurityScorecard
An LLM chatbot over security data for 12M+ companies, in production 6 weeks after starting from zero
The Problem
SecurityScorecard rates the security of 12M+ companies. The data behind those ratings is complex. Non-technical users struggled to work with it, which limited how much of the platform they actually used.
Constraints
- Built from zero, with 6 weeks to production
- A security-critical product, so the chatbot had to be reliable, not a demo
- At the same time I was the sole backend owner of three Django systems: the platform API, the support portal and the intranet
What I Did
I led the full backend of the chatbot.
- Orchestrated the conversation and data retrieval with LangChain
- Used open-source HuggingFace models
- Built the serving layer with FastAPI
- Added context-aware query processing and kept session state, so users could ask follow-up questions
- Deployed it to AWS on Kubernetes
Result
6
Weeks to Production
12M+
Companies Queryable
Users can ask questions in plain language about security score data for 12M+ companies. Business users who could not work with the raw data now get answers directly.
Stack
Python
LangChain
HuggingFace
FastAPI
Snowflake
AWS
Kubernetes
Want an LLM Feature in Your Product?
Tell me what users need to ask and where the data lives. See how I work on AI integration.