Developed necessary features during AI platform development, bringing RAG knowledge service, MCP Proxy, and DB access control engine to production level in a short time.
AI PlatformSecurityRapid Delivery
Acted as both PL and IC, bringing knowledge-based RAG service to first deployment in 1 month 3 weeks.
Highlights
- Designed and implemented entire upload → parsing → chunking → embedding pipeline.
- Introduced chunking-based streaming indexing in 2 days, reducing memory by 90% for text and 40% for PDFs, processing 10MB files with 1Gi resources.
- Encrypted customer knowledge across all storage stages with CBC encryption and conducted real-time profiling with py-spy/memray + ephemeral container.
- Managed retriever quality with DeepEval metrics, achieving 80% correctness on 100 QA pairs.
Tech Stack
PythonFastAPICeleryKotlinSpringPostgreSQLKubernetes
Pioneered development of MCP protocol-interpreting Proxy and internal work agents/registry, currently patent pending.
Highlights
- Completed MCP Proxy for tool access control and audit logs in 4 days, demonstrated at company town hall.
- Implemented multi-log analysis and meeting preparation automation with ADK-based ErrorSherlock and BriefButler agents.
- Built Registry server/admin/CLI and controlled MCP processes via IPC, configured so LLM suggests execution scripts when only GitHub URL is input.
Tech Stack
TypeScriptNode.jsPythonADKSvelte
Maintained DB Proxy/Engine layer and expanded new protocols.
Highlights
- Added RESP3 (redis) protocol across entire Proxy and Engine sections.
- Conducted Oracle protocol reverse engineering to define future support scope.
- Checked QueryPie user permission verification and masking logic, improving operational quality.
Tech Stack
C#.NETRedisProtocol Engineering
