Portfolio
Some architecture diagrams and code snippets are abbreviated for clarity; you can view the full version with better readability at link.
Sionic AI
2024.09 — now- →
OpenGateway Development/Operations
API Gateway · Model Routing · Prompt Cache · Spring · Kotlin · Billing · Multi-deployment- Extended an internal model routing engine into an OpenAI-compatible public API Gateway product
- Serving RPM 180, Daily 250K, 10+ Providers, 100+ Models
- →
Storm Parse APIs Development
Spring · Kotlin · Multi-deployment- Extended an internal document-parsing feature into a public API product that external customers can integrate directly
- Multi-deployment support, drove the first external SaaS revenue
- →
Search Data Serving Architecture Improvement for Bank A
Qdrant · Parquet · Performance- Validated stability and improved performance of a search data serving and indexing pipeline at 7.1M-vector scale
- Reduced Indexer processing time from 3 hours 30 minutes to 24 minutes (88.6% reduction)
- →
Operations Support Tooling and Workflow Standardization
OpenAPI · Apidog · BO · Python · Node.js- Standardized documentation and testing flows with an in-house OpenAPI-based API Hub
- Strengthened monitoring with regression tests and AI reports on WARN logs
- Developed operations support tools such as onepage-payment, storm-differ, and BO
- →
Thoughts on Development That Actively Leverages AI
- Separating the flows humans must control from the areas that can be delegated to AI
- Harness engineering that manages policies, research, and Skills as a single source of truth
Wrtn Technologies
2024.01 — 2024.07- →
AI Model Serving Service Development & Operations
Spring AI · JVM · Kafka · Datadog · Gatling- Developed the JVM-based hamlet2 as an extension of Spring AI and replaced the existing Node-based hamlet1 with zero downtime
- Achieved RPM 750 and a 20% improvement in API response time
- →
Company-wide Common Service (CDS) Development & Operations
Python · FastAPI · Tiktoken- Provided functionality shared across multiple services through an API server (tokenizer, email validation, profanity detection, etc.)
- The following describes the tokenizer, its representative feature
Kurly Pay
2022.09 — 2023.12- →
Kurly Pay Gift Card Design, Development, and Operations
Spring · Kafka · B2B · B2C · Payment- Supported 4 use cases (B2C/B2B/corporate card/individual bulk purchase), separated external-partner and Kurly Mall flows, and handled the gift card lifecycle via Kafka
- →
Improving the Kurly Mall Gift Card Purchase Process
Kafka · Eventual Consistency · Migration- Redesigned data inconsistencies during failures into a Kafka-based eventually consistent structure that could self-recover, connecting the Go legacy with the new Kotlin/Spring flow
- →
Building Statistics Monitoring and Researching Scalability
Redash · Monitoring · Statistics- Built Redash-based dashboards and researched a scalable statistics architecture
Troubleshooting
-
OutOfMemoryError During Batch Jobs
→With the owner having left the company, applied a short-term mitigation for batch OOM and drove a company-wide memory optimization.
-
Instances Failing to Start After Deployment
→Resolved intermittent startup failures by separating the responsibilities of startupProbe and livenessProbe.
Essays
- →
How I Work
- Three practices I rely on to stay effective — making problems explicit before delegating, turning individual knowledge into team assets, and closing the loop between learning and production