Instrumented all AI pillar microservices with OpenTelemetry, enabling end-to-end distributed tracing and reducing mean time to debug production incidents by 40%.

• Optimized DLP pipeline and OTel Collector in Go using Google Cloud DLP table format, cutting latency by 74% and eliminating span loss.
• Built a Go trace processor that extracts agent metrics (inputs, outputs, intent, tool calls) from OTel spans into BigQuery. Cut data scientist query turnaround by 53%.
• Architected a Google Pub/Sub feedback loop for real-time customer feedback ingestion, letting analysts act on insights without manual data pulls.
• Integrated Grafana and Arize across AI microservices, centralizing observability for service monitoring and model evaluation.

Researched and developed a 4-input, 4-output audio device on a microcontroller with real-time DSP capabilities.

• Designed a system on a dual-processor microcontroller to offload DSP and console workloads.
• Engineered DSP algorithms in JUCE/C++, optimized via decimation and FFT to cut processing latency by 25%.
• Tuned Linux kernel parameters and used DMA optimizations to bring audio latency below 5 ms.
• Built FreeRTOS-based IPC between the dual processors and the host system.

Built an SMS-based application in Java/Spring enabling LLM access for users without internet connectivity.

• Integrated LiteLLM to unify access across multiple LLMs, reducing model-switching latency by 56%.
• Integrated the Twilio API with delivery callbacks and auto-retries to deliver alerts in under 2 seconds.
• Shortened setup time by 70% by containerizing with Docker and automating deployment on GCP.