Projects

Open-source, open-core, and selected closed-source work in battery systems, embedded systems, and safety-critical applications.

celljar viewer - interactive explorer for harmonized battery test data
Released Open Data Battery Testing Canonical Schema Parquet + SQL HuggingFace

celljar

Public battery test data is scattered across many sources, each with its own format, units, and structure - so every project starts with custom wrangling before any analysis.

celljar harmonizes published datasets (10 so far, spanning 8 cell models / 280 cells / ~184M time-series rows) into one canonical schema: metadata as JSON, time series as Parquet for fast queries and compact storage. Query across every dataset with simple SQL / pandas / Polars, or explore them in the included viewer. Free and hosted on HuggingFace; code on GitHub.

ess_pysim - 3S2P pack constant-power discharge: terminal voltage vs OCV, plant vs BMS-estimated SOC, and cell temperature
Beta BMS Algorithms ECM Cell Model Parameter Extraction SOC / SOH / SOP / SOE Python

ess_pysim

BMS estimators are only as good as the cell model behind them, and the model is only as good as the data its parameters were fitted from - accuracy shown on one cell, temperature, or drive profile does not automatically carry to another.

A Python framework, currently built on public datasets and test protocols, that closes the loop from raw data to validated BMS algorithms: cell parameters from celljar's harmonized datasets drive electro-thermal ECM models and a full BMS application layer, from cell to pack. Validated out-of-sample against UDDS and US06 drive cycles (26 mV RMSE, Samsung 30T at 25C). Request a demo.

TraceStack functional decomposition view - functions with development assurance levels and allocated components
Alpha ARP 4754B ARP 4761A DO-178C ISO 26262 FastAPI + React

TraceStack

For safety-critical systems, requirements anchor every phase gate, from SRR through certification. But requirements, safety analyses, HW/SW artifacts, and their V&V live in disconnected tools and dashboards, so change impact is traced by hand and certification/compliance evidence is rebuilt before every review.

A single platform for the whole assurance thread - requirements, trace links, safety analyses, and verification evidence - where a change on either end auto-flags what it touches, so impact is surfaced rather than hunted and review evidence stays current. Aligned to ARP 4754B / ARP 4761A / DO-178C. Early alpha - demo available on request.

telemetry-specwatch - DCIR trend monitored against its end-of-life spec on the edge, drift flagged with predicted spec crossing ~40 cycles ahead
Concept Intelligent Telemetry Edge AI NVIDIA Jetson Anomaly Detection

telemetry-specwatch

Spec limits live in documents while telemetry streams past unchecked - excursions surface in post-processing, long after the test article has moved on.

SpecWatch turns specifications into live, on-edge monitors (NVIDIA Jetson): telemetry is checked in real time against spec-derived envelopes, so excursions and drift raise alerts as they happen rather than in review. Under conceptual development.