Data Scientist for TensorFlow Lite Model Development on ESP32

Firmware Engineer (ESP32-S3 • TFLite Micro • Session storage • BLE/Wi-Fi)
Context: BetterE is a universal charge limit for e-mobility devices, using advanced ML models, it accurately estimated the state of charge of batteries without DC or BMS readings, on the AC-side of the charger.
You will be helping develop the firmware. Core I/O is done (Energy metering chip reading from the ESP, relay driving, LEDs/buttons working). We now need a pro to make the firmware smart and production-ready.
What you’ll do
• Session storage: design and implement on-device storage of past charge/usage sessions (start/stop timestamps, energy, peaks, flags) with loss-safe writes and compaction (NVS / SPIFFS / LittleFS).
• On-device ML: run a TensorFlow Lite Micro model on ESP32-S3 (quantized). Handle model load, inference loop, memory planning (arena sizing), and versioning.
• Feature pipeline: compute the model inputs from both realtime energy metering telemetry and stored history. Include windowing, normalization, derived features, and (if needed) a secondary lightweight model/transform stage.
• UX on-device: make buttons/LEDs intuitive (short/long/very-long press mappings, state machine, safe relay control, feedback patterns).
• Connectivity: add stable Wi-Fi + BLE paths. Push telemetry + session summaries to cloud (MQTT or HTTPS REST). Handle reconnect/backoff, buffering, and retry after power loss.
• Config & OTA: simple settings store (SSID, tokens, thresholds), and OTA updates.
• Quality: watchdogs, metrics/logging, and a small suite of unit/integration tests (FreeRTOS friendly).
Must-haves
• Strong C/C++ on ESP32 (ESP-IDF or Arduino-ESP32 with IDF features), FreeRTOS tasks/queues/timers.
• Hands-on with TFLite Micro on MCUs (quantization, tensor arena, perf trade-offs).
• Durable embedded storage patterns (journaling, ring buffers, GC/compaction).
• BLE (GATT) and Wi-Fi flows on ESP32; secure cloud comms (TLS), basic cert/token handling.
• Comfortable turning informal feature specs into clean state machines and tests.
Nice-to-haves
• Experience with power/energy devices (HLW/ADE/ATM90xx), calibration, and windowed stats.
• MQTT at scale (QoS, retained messages), CBOR/Protobuf payloads.
• OTA via HTTPS, version gates/rollback.
• Basic analytics on-device (EWMA, percentile, peak detection) to enrich ML features.

• Develop TensorFlow Lite models for ESP32
• Implement AC charge power curve recognition
• Optimize models for performance on ESP32