Ravindran S Logo
Ravindran S
Engineering Portfolio

Featured Projects

Deep dives into sub-1 GFLOP computer vision models, kernel-level eBPF package security in Rust, and real-time distributed IoT telemetry.

Edge AI & MLFeatured

MicroDet

Sub-1 GFLOP Real-Time Drone Detection Pipeline

Metric 1

<1.0 GFLOP Budget

Metric 2

96% mAP@0.5

Metric 3

30+ FPS INT8 Inference

Metric 4

vs 89% MobileNetV2 Baseline

Architectural Breakdown

Architected an ultra-lightweight object detection neural network designed specifically for compute-constrained drone hardware payloads.
Coupled an optimized NanoDet backbone with custom YOLO decoupled heads, achieving 96% mAP@0.5 on target benchmarks.
Applied TensorRT INT8 calibration and quantization, delivering 30+ FPS sustained on embedded accelerators, outperforming standard MobileNetV2-SSD baselines (~89% mAP) at equivalent FLOP budgets.
Presented and demonstrated at NIDAR 2025 autonomous robotics symposium.
Built With:TinyGradPyTorchTensorRT INT8DeepStream SDKCUDAC++Python
Systems & SecurityFeatured

Archon

Security-Hardened AUR Package Manager with eBPF Sandboxing

Metric 1

<5% Build-Time Overhead

Metric 2

Kernel-level eBPF Probes

Metric 3

Bubblewrap Isolated Jail

Metric 4

Libalpm Graph Resolver

Architectural Breakdown

Engineered a high-performance, security-focused AUR (Arch User Repository) package manager in Rust.
Implements dynamic dependency graph resolution reconciling official Arch repositories and untrusted AUR PKGBUILDs.
Sandboxes unverified compilation scripts via Bubblewrap and attaches kernel-level eBPF probes (using Aya) to monitor syscalls (execve, openat, socket connect) in real-time.
Detects malicious script injections and unauthorized exfiltration attempts with under 5% build-time overhead.
Built With:RusteBPF (Aya)BubblewraplibalpmLinux KernelSecurity Systems
IoT & Full-StackFeatured

Machine Guard

End-to-End Industrial IoT Predictive Maintenance Pipeline

Metric 1

<200ms Alert Latency

Metric 2

94% Anomaly Accuracy

Metric 3

Rs. 50,000 Hackathon Winner

Metric 4

Real-Time MQTT Stream

Architectural Breakdown

Architected a distributed IoT telemetry pipeline for industrial machine vibration and acoustic anomaly detection.
Embedded quantized TFLite models on edge sensor nodes achieving 94% anomaly detection accuracy.
Streamed high-frequency sensor readings over MQTT to Firebase and synchronized to a high-density React operational dashboard with <200ms end-to-end alert latency.
Official winning submission at Ti Forge 2026 Hackathon (awarded Rs. 50,000 cash prize).
Built With:KotlinPythonTensorFlow LiteMQTTFirebaseReact.jsIoT Sensors

Looking for more repositories and experimental code?

Check out open-source toolkits, algorithms, and systems experiments on GitHub.

github.com/ravindran-dev