Embedded / System Engineer
Cross-layer HW/SW Debug • Automation • AI-assisted Diagnostics
I turn ambiguous field failures into root causes — and manual diagnostics into automated pipelines.
Deep-dive investigations across hardware, firmware, Linux stack, and storage. Systematic approach from symptom collection through hypothesis testing to definitive root cause — turning NFF-style ambiguity into answers.
Building multi-agent pipelines with tool use (MCP) for automated evidence collection, hypothesis generation, and retrospective learning. Python automation for snapshot parsing, KPI extraction, and repeatable reporting. Engineer stays in the loop as final arbiter — raw logs are always ground truth.
From schematic capture through PCB layout to hardware bring-up. Embedded Linux drivers, firmware, and low-level interface integration. IoT connectivity with BLE/Zephyr, Zigbee, Z-Wave.
Comprehensive investigation into EXT4 filesystem corruption in field-deployed telecom units. Validated that power-loss during metadata updates — not SSD hardware failure — caused corruption patterns. Analyzed Hyperstone U8 vs U9 controllers and capacitor discharge timing in Swissbit SSDs.
Software-based approaches for identifying cold solder joints in deployed radio cards. Correlated reset cycles during software upgrades with joint degradation — creating predictive workarounds to reduce ECC errors in live telecom equipment.
IoT gateway concept for home/edge scenarios. End-to-end connectivity solution demonstrating protocol bridging and sensor data aggregation for smart environment monitoring.
nRF52-based sensor nodes with e-ink display, Zephyr RTOS firmware. BME680/SHTC3 environmental sensing with motion detection. CR2032 battery-focused power optimization for months of autonomous operation.
Agentic pipeline for No-Fault-Found investigations in telecom RAN hardware. Specialized agents handle evidence collection, cross-layer log interpretation, minimal-evidence triage, and retrospective learning from past cases. Connected via MCP server to live diagnostic tooling. Reduces time-to-hypothesis on complex intermittent failures while keeping the engineer as final decision-maker.
Two-level diagnostic system: deterministic scripts handle known failure patterns; specialized AI agents tackle unknown failure modes. Agents are connected to live diagnostic tools via MCP, enabling automated evidence collection and hypothesis generation. Multiple agents handle distinct investigation phases — interpreter, minimal-evidence triage, retrospective learning — with the engineer verifying conclusions against raw data.
Always happy to talk about telecom diagnostics, embedded systems, and agentic AI.