Open to Full-Time Roles · Graduating 2027

I build things

Final-year Electronics and Communications Engineering student at VIT-AP, graduating 2027. I build embedded firmware, Linux drivers, and hardware accelerators. Most problems get interesting when you stop trusting the abstraction.

Atul Akella
Currently on workbench
Embedded Linux | Firmware Dev
Scroll
01 About Me

I go lower
until things make sense.

It started with a bluetooth robotic arm at 13. I didn't have a datasheet. I didn't know datasheets existed. All I knew was I wanted to understand why it worked.

That question "why?" never left. It kept following me through every project. And every time I thought I had the answer, I'd realize there was another layer worth understanding.

Now I enjoy working at the hardware-software boundary. I'm drawn to problems that reward understanding the system beneath the abstraction, whether that means debugging, designing, or building from the ground up.

Outside of that I play a lot of chess, occasionally basketball, and spend more time than I should listening to a song on repeat until I've figured out exactly what the lyrics are doing.

Firmware & OS
Bare-metal C FreeRTOS OpenAMP RPMsg Embedded Linux ARM Cortex-M4 / A7
Hardware & FPGA
Vitis HLS Vivado AXI4 Zynq Z7-10 Fixed-point arithmetic PS+PL co-design
Wireless & Drones
ESP-NOW MAVLink CRSF iNav UART / SPI / I2C
ML at the Edge
TinyML YOLOv8 DETR SAM On-device inference
02 Experience

Where I've worked.

May – Jul 2025
NIT Rourkela | Rourkela, India
FSO | Optical Comms Hardware
  • Built a simplex FSO data link from scratch: custom UART framing with byte stuffing, NRZ encoding, and multi-protocol input bridging across I2C, UART, SPI, RJ45, and RS232 on a single transmitter node.
  • Achieved reliable text and image transmission over a free-space laser link. Signal recovery pipeline built directly on raw photodetector output with no off-the-shelf modem layer.
  • Designed the full-duplex extension path: link-layer ACK/NAK retransmission, framing protocol, and end-to-end testing over actual free-space propagation distance.
Sep 2023 – Dec 2024
ShrimpIQ | Remote
Edge ML | Computer Vision
  • Benchmarked YOLOv8-nano, DETR, and SAM on aquaculture field data for shrimp defect detection. Achieved +6% mAP over the YOLOv8-nano baseline on real production images.
  • Optimized the classification pipeline for edge deployment: quantization, pruning, and operator fusion. Validated across ARM Cortex-A and MCU-class targets on real field data.
03 Projects

Things I've built.

Hardware, firmware, and ML. Sometimes all three at once.

Jun 2026
Real-Tim Shady

Full heterogeneous SoC bring-up from scratch on the STM32MP157D. Custom Yocto BSP layer (meta-motorcontrol, Scarthgap), PREEMPT_RT Linux 6.6.116-rt66 on the Cortex-A7, FreeRTOS 1kHz PID loop on the Cortex-M4, bidirectional RPMsg IPC over OpenAMP, and a custom kernel module exposing /dev/daq0 (telemetry) and /dev/daq1 (gain commands). IPC latency characterized across 80,000 samples and 8 load conditions: mean 58us, P99 grows 57x under CPU saturation, governor-invariant, payload-size invariant up to 256B.

STM32MP157D Yocto BSP PREEMPT_RT Linux 6.6 FreeRTOS 1kHz Custom kernel module P99 57x under load 80K samples
Apr 2026
CNN Inference Accelerator

Custom convolution accelerator on a Zynq-7000, using Vitis HLS and AXI4 streaming between PS and PL. I wrote the HLS kernel for the convolution layer, optimized the dataflow and loop unrolling until throughput was 2.67x that of a software-only baseline running on the Cortex-A9 PS, at 1421 inferences/sec, while consuming 5x fewer DSP48 blocks than an unoptimized HLS reference design.

Zynq-7000 FPGA 2.67x speedup 1421 inferences/sec 5x fewer DSPs
2024
Electronic Nose

Triggered by the Tirupati laddu adulteration controversy: could a small device tell the difference between pure and adulterated ghee by smell alone, with no lab equipment? Built a sensor array that reads the chemical signature of heated ghee, trained a tiny classifier that fits on a microcontroller, and got to 86.6% accuracy running entirely on-device.

86.6% accuracy Runs on microcontroller 15.8 KB flash 1 ms inference
Mar 2026
Tethered Drone

A drone you control over a cable, for situations where radio links are unreliable or battery life matters more than freedom of movement. I replaced the standard remote control radio with a custom communication stack, built a flight dashboard with live telemetry, and added a circuit to switch between tether power and battery mid-flight. It flew.

Custom radio stack Live telemetry dashboard Mid-flight power switching iNav failsafe confirmed
2024
VISU-NG, Humanoid Robot

My university's humanoid robot. It could hold a conversation, recognise faces and remember who it had met before. Demonstrated to the HRD Minister of Andhra Pradesh at two public events.

Voice interaction Face recognition Per-person memory Public demo
More on GitHub

Experiments, side projects, half-finished ideas, and things that didn't make this list. The interesting stuff usually lives here first.

04 Achievements

A few things
worth noting.

Age 13

Built my first robot

Bluetooth-controlled robotic arm, before I owned a multimeter or knew what a datasheet was. The start of everything.

Age 14

IIT Kharagpur, top 30 of 3000+ teams

Led my school team to the finals of the pan-Asia Young Innovators Program (YIP 2019). First time competing at that scale.

Age 15

Built my first portfolio

Designed and shipped a personal site with an interactive chatbot. First real full-stack project, and the reason this one exists.

Age 19

Three hackathons, college freshman year

National Hackathon 2024 (VIT-AP) and TinkerQuest '24 (IIT Roorkee × Redcliffe Labs), among others. Rapid iteration under pressure turned out to be a useful skill.

Age 20

Two IEEE publications

First-authored two IEEE conference papers on embedded ML and hardware-software co-design. Peer-reviewed, cited work in the embedded systems research community.

Age 20

₹14L MSME Grant, Govt. of India

Government innovation grant for CogniLift, a neurofeedback-based brain activity monitoring and therapy device. First time building something at the brain-computer interface, and the first time someone external put money behind the direction.