June 2025 - Present
Software Engineer
CurrentSansay Inc./San Diego, CA
Sole engineer behind Sansay’s AI support agent, used in production by customers and internal support engineers, and a telecom fraud-detection system currently in development.
- median call processing, JS workers to Go
- 977 → 61 ms
- median call processing, JS workers to Go
- call records aggregated per month
- ~500M
- call records aggregated per month
- technical docs in the RAG index
- 800+
- technical docs in the RAG index
AI Support Agent
- Built and deployed a production AI support agent in TypeScript, using LangGraph for agentic orchestration and self-hosted LLMs served with vLLM, used to diagnose telecom issues.
- Integrated the agent into VSXi, Sansay’s main product, using Java 8 Servlets.
- Designed a tool layer of read-only tools that query live customer systems, replacing manual checks and documentation searches during troubleshooting.
- Built a hybrid RAG pipeline over 800+ technical documents and thousands of past support tickets, combining full-text search, vector search, and a reranking model.
- Automated SIP Message Conversion profile creation, replacing a multi-hour manual process that support ran several times a week.
Fraud Detection System
In development- Architecting a telecom fraud-detection system in Go that scores live call metadata against aggregated historical records and exposes results through a REST API.
- Built a pipeline that aggregates ~500M call records per month by customer and carrier to drive false answer supervision (FAS) detection and anomaly scoring.
- Rewrote call processing from JavaScript workers to Go goroutines, cutting median processing time from 977 ms to 61 ms on an 80,000-active-call benchmark.
- TypeScript
- LangGraph
- vLLM
- Go
- Java
- RAG
