Backend systems built for scale, speed, and recovery.

Agentic AI Engineer Intern · AIMING Lab

• Co-built core components of an end-to-end multi-agent AI research system, orchestrating a 23-stage pipeline from research question to conference-ready paper; the open-source project reached 14.5k GitHub stars.
• Designed a human-in-the-loop ablation across five intervention strategies, identifying six high-stakes decision gates that increased agent evaluation scores from 4.0 to 7.3 while preserving autonomous execution.
• Built a fault-tolerant multi-agent recovery policy with checkpointing and REFINE / REDESIGN / PIVOT paths, enabling stage-level recovery instead of full-pipeline restarts and saving ~$2K per failure in redundant LLM calls.

• Python
• Claude Code
• OpenAI Codex
• Agent orchestration
• YAML / JSON

Backend Software Engineer Intern · Synergies Intelligence

Refactored a legacy monolith into Java Spring Boot microservices, improving latency by 33%. Built a Kafka event pipeline that removed 30% of billing delay and added 2,000 RPS, then redesigned PostgreSQL schemas and indexes to reduce duplicate-record errors by 95%.

• Java
• Spring Boot
• Kafka
• PostgreSQL
• Microservices

C++ Software Engineer Intern · UNC Computer Science

Led post-training quantization for a closed-loop reinforcement-learning controller deployed to ESP32-S3. Achieved 4× faster response and 27% lower power use, then removed a 32 KB cache bottleneck with configurable C++ tiling to cut dense-layer latency by approximately 62%.

• C++
• PyTorch
• TensorFlow
• ESP-IDF
• Embedded ML

Backend Software Engineer Intern · Auneau Tech Co.

Improved a Spark and Airflow ETL pipeline handling 1.4 TB of logs per day, reducing runtime by 28%. Accelerated portfolio APIs for 20,000 real-time users with Redis and AWS Lambda validation, lowering P95 latency by 700 ms and malformed-record incidents by 75%.

• Spark
• Airflow
• Redis
• AWS Lambda
• Data validation