Hyderabad, India/Open to opportunities

KAMAL BURA

AspiringEmbedded AI Engineer

Aspiring Embedded AI engineer in Hyderabad. I build real-time systems that run AI on the device — the firmware keeps the deadlines, the model does the thinking.

B.E. CSE (AI&ML), Vasavi College of Engineering '26

See the builds View résumé

Scroll

GitHub LinkedIn Email

About Me

Crafting Intelligent
Solutions for Tomorrow

Aspiring Embedded AI engineer in Hyderabad. I build real-time systems that run AI on the device — the firmware keeps the deadlines, the model does the thinking.

The projects below are the argument: robots you can talk to, radios I built instead of bought, models running on hardware I can hold, and security research for UAVs.

Education

B.E., Computer Science & Engineering

AI & Machine Learning · Vasavi College of Engineering, Hyderabad

Graduated June 2026 · CGPA 7.6

GATE 2026 (Computer Science) — Qualified, AIR 46071, Score 298

Core stack

ESP32 / ESP32-S3C / C++ESP-NOWRaspberry PiPythonLinux

5
Hardware projects shipped: 99%
Edge server uptime, 4 months: 3
Engineering roles since 2025: ♾️
Curiosity

Kamal Bura at the Vasavi College of Engineering graduation ceremony

Technical Constellation

The Stack That
Powers Innovation

ControlLinkAI & Compute

|Hover a skill for context

C / C++
ESP32 / Microcontrollers
FreeRTOS / RTOS
PID Control & Motors
I2C / SPI / UART
Sensor Fusion & Kalman
ESP-NOW & LoRa
MQTT & IoT Protocols
GSM / Cellular
Wi-Fi & Network Security
Embedded Linux
Hardware Debugging
Edge AI / TinyML
PCB Design Basics

Portfolio

Featured Projects

Robotics / Perception01 / 08

Voice-Controlled Robotic Vehicle

A robotic vehicle you talk to. Wake word, speech-to-text, an LLM parsing the command into drive instructions — and an ultrasonic safety layer that overrides all of it.

Two-tier split: the Pi handles perception and language, the ESP32 holds the real-time control deadline
Ultrasonic obstacle-stop overrides any voice command — it refuses to drive into things
ChaCha20-Poly1305 encrypted control link between Pi and ESP32
ZeroMQ IPC across the audio, language, and control services; live FFmpeg camera stream
YOLOv9 via ONNX Runtime; spatial navigation to spoken objects in active development

Raspberry PiESP32PicovoiceAzure STTGeminiZeroMQYOLOv9 / ONNXFFmpegChaCha20-Poly1305

Working — spatial navigation in progressCode

Wireless / Motor Control02 / 08

Competition RC Vehicle & Custom ESP-NOW Radio

A national-competition RC monster truck where even the radio is custom — a dual ESP32-S3 ESP-NOW link replaces the off-the-shelf transmitter entirely.

Handheld ESP32-S3 transmitter sends throttle and steering over ESP-NOW with a 3dBi antenna
Onboard ESP32-S3 receiver drives four geared DC motors through BTS7960 H-bridges in differential steering
Custom pairing flow, packet-loss handling, and failsafe — motors stop on link loss
Custom-built 3S LiPo pack sized for race current draw
The step from assembling an RC car to engineering the RC system itself

ESP32-S3ESP-NOWBTS7960PWM3S LiPoDifferential Steering

Built and racedCode

MLOps / Time Series03 / 08

Air Quality Forecasting Pipeline

End-to-end Air Quality platform running on a Raspberry Pi: Open-Meteo ingestion, PostgreSQL with watermarks, per-city Airflow DAGs, and a 9-model zoo tuned with Optuna.

Incremental load watermarking into PostgreSQL ensures interrupted API pulls resume cleanly without duplicates
Airflow orchestration isolating per-city DAGs so one API failure doesn't stall others
Model factory pattern spanning 9 architectures (XGBoost, Transformers, LSTMs) with automated Optuna tuning
Serving layer built in Next.js pulling from the shared database

PythonPostgreSQLApache AirflowXGBoostNext.jsOptuna

Working pipelineCode

Security / Edge ML04 / 08

DDoS Detection on Raspberry Pi

Two mutually exclusive DDoS detectors (XGBoost and INT8-quantised Transformer) built for the thermal and latency constraints of a Raspberry Pi.

Shared feature extraction pipeline means switching from XGBoost to Transformer is just a launcher config change
Evaluated thermal throttling as a primary constraint for edge ML — benchmarks measure sustained heat, not just burst latency
Verified INT8 quantisation did not degrade the decision boundary on the long tail

XGBoostTransformerINT8 QuantisationEdge ComputingPython

Complete

IoT / Firmware05 / 08

DeskMate

An always-on ESP32-S3 desk companion with an animated expressive face, gesture reactions via MPU6050, and OTA updates.

Gesture as the only input — tight MPU6050 orientation thresholds prevent desk-bump triggers
Strict idle sleep budgets power down the display after 40 seconds of stillness
OTA (Over-the-Air) firmware updates enable iterative development without a cable

ESP32-S3C++MPU6050OLEDOTASPI

BuiltCode

Real-Time Control06 / 08

ESP32 Flight Controller

Flight-stabilization firmware from first principles: raw IMU readings, sensor fusion, PID loop — built around loop timing rather than code convenience.

MPU6050 accelerometer and gyroscope over I2C
Hybrid Kalman/complementary filter — gyro for fast response, accelerometer to correct drift
PID loop computing stabilization corrections against the fused orientation estimate
Optimized for latency: a stabilization loop is only as good as how fast it closes

ESP32MPU6050I2CKalman FilterPID Control

Working firmware, bench-tested

Embedded / Wireless07 / 08

GNSS/GSM Safety Alert Device

Press one button and it texts your live GPS location over the cellular network. No smartphone, no Wi-Fi, no app.

NEO-6M GPS streaming NMEA over UART, parsed with TinyGPS++ into live coordinates
SIM800L GSM sends an SMS with the location to preset contacts on a button trigger
Designed for independence — works anywhere with cellular coverage, no paired phone

NEO-6M GPSSIM800LUARTTinyGPS++OLED

Working prototypeCode

Infrastructure / Linux08 / 08

Self-Hosted Edge Infrastructure

A Raspberry Pi 4 run like real infrastructure rather than a hobby box — DNS filtering, private VPN mesh, NAS, and full metrics, with uptime it is accountable for.

99% uptime across four months of continuous operation (Feb—Jun 2026)
Dockerized Pi-hole, Tailscale, Samba NAS, and Grafana + Prometheus
Stable 2GHz overclock, validated at 100% CPU for 5 minutes under 60°C

Raspberry Pi 4DockerPi-holeTailscaleSambaGrafana

In continuous operationLive Code

Journey

Professional Experience

Building expertise through impactful work

AI Engineer & Data Engineer
Saras Analytics
Jan — Aug 2026
Built a multi-agent QA orchestration system for data pipelines — currently in testing ahead of production rollout.
- Designed agent skills as custom skill files: grounding, categorization, and RCA generation
- Agent routing hands ETL/Airflow-specific gaps to a dedicated data-engineering agent
- Validated correctness, consistency, and completeness of production datasets
Freelance IoT Developer
Independent Client Projects
Sep — Dec 2025
Delivered IoT prototypes end to end: sensor integration, microcontroller firmware, and real-time data processing.
- Client-facing debugging, calibration, and iteration cycles
- Built deployment discipline on hardware that had to work outside the bench
IoT Research Intern
IIIT Hyderabad
May — Aug 2025
Worked on security systems on UAVs, hardware debugging, and drones.
- Built and integrated security systems on edge hardware
- Performed extensive hardware debugging for drone control links
- Tested and validated communication links on custom-built UAVs