Data Scientist for Physical Systems · Embedded Control & Estimation I turn sensor and control-loop data into decisions and I can fix the data at its source.
MSc (c) Data Analytics · Electronic Engineer · Bogotá, Colombia · Jitter Ingeniería SAS
The most valuable industrial data comes from sensors, drives and control loops and it is rarely clean. Drift, noise, irregular sampling and controllers that mask faults break models long before the modeling starts. Most data scientists cannot see those problems; most electronic engineers do not model them. I work on both sides.
| Layer | What I bring |
|---|---|
| Data analytics | Forecasting, anomaly and fault detection, uncertainty quantification, out-of-sample validation, prescriptive analysis |
| Estimation | Kalman observers and residual-based identification of slowly drifting physical parameters winding resistance, back-EMF constant, sensor drift |
| Data at the source | Embedded firmware, real-time control and instrumentation on STM32 and AVR, so the data is right before it reaches a model |
Where this applies: predictive maintenance and condition monitoring · industrial IoT · power quality and energy systems · validation of electronic products.
Learning-based control that keeps its guarantees after deployment. Neural and data-driven controllers are usually certified in idealized arithmetic and continuous time. I study what remains of those guarantees on a real microcontroller fixed-point arithmetic, finite execution time, sampling jitter and validate on physical hardware, not only in simulation.
dc-motor-fault-detection · in progress
Fault dataset and detection methods for DC drives operating inside a closed control loop, with severity in physical units. Public datasets are usually recorded open-loop with categorical labels; in a real loop, the controller masks incipient faults and the signature moves into the control effort. This platform captures exactly that, on an STM32F407VG testbed built for the purpose.
Python · scikit-learn · Kalman filter · STM32 HAL · C · DMA telemetry · PyQt6
Can low-cost sensor drift be estimated from residuals against neighboring nodes and a reference station, without periodic physical recalibration? A station in continuous field operation in Bogotá since September 2026, with an analysis chain covering quality control, time-scale separation, out-of-sample forecasting and operational prescription.
Python · QC pipelines · forecasting · embedded firmware · CI
Industrial thermal controller for 120 VAC LED panels: PID regulation through phase-angle fan modulation, zero-cross self-calibration with 50/60 Hz auto-detection, eight fault conditions with operator acknowledgment, CSV telemetry with Python tuning tools, and a hardware thermal failsafe independent of the microcontroller.
C++ · AC phase control · EEPROM/CRC-16 · Python · open hardware
- Physical units over labels. A threshold that cannot be expressed in ohms, volts or degrees does not transfer to another site.
- Claims match evidence. Every project separates what the instrument measures from what the data allows us to conclude, and documents its known limitations.
- Safety lives in hardware. Firmware can fail; the last line of protection should not depend on it.
| Domain | Tools |
|---|---|
| ML & data | Python, PyTorch, scikit-learn, pandas, Plotly, uv |
| MLOps | MLflow, Delta Lake, Databricks, GitHub Actions |
| Estimation & control | Kalman filtering, PI/PID, system identification |
| Embedded | C/C++, STM32CubeIDE, HAL, DMA, Arduino, PyQt6 |
| Instrumentation | Oscilloscope, logic analyzer, serial telemetry, custom PCBs |
- MSc (c) Data Analytics, Universidad Central de Colombia
- BSc Electronic Engineering
- 8 years in IP/MPLS network engineering for telecom operators — carrier-grade production systems
- Engineering lead on client projects at Jitter Ingeniería SAS
Open to: industrial data science roles in predictive maintenance, condition monitoring and IIoT · research collaborations and doctoral positions in learning-based and embedded control.