Shyam Sreenivasan

Apr 2026

Stereo Camera Health Monitor for ORB-SLAM3

Course: EECE 5554 Robot Sensing and Navigation, Northeastern University

• Built an end-to-end SLAM health monitoring pipeline on TUM-VI — sweeping five image degradation types (Gaussian blur, motion blur, salt-and-pepper, brightness, occlusion) across ORB-SLAM3 to characterize ATE failure thresholds
• Trained a ResNet18+GRU model on 20-frame stereo windows with self-supervised severity labels derived from SLAM trajectory outputs; severity metric triggers automatic switch to mono-inertial mode at degradation threshold
• Results: Val MAE 0.135, 80% crossover classification accuracy (65 windows, 13 conditions); key finding: blur produces learnable warning zones while salt-and-pepper and occlusion produce binary cliffs absorbed by RANSAC

View on GitHub →

Apr 2026

Deep RL Policy Training: PPO, A2C, DQN with Reward Shaping

Course: CS 5180 Reinforcement Learning, Northeastern University

• Implemented PPO, A2C, and DQN from scratch in PyTorch with GAE, entropy regularization, and clipped surrogate objectives; ablated four reward shaping strategies (hyperbolic, threshold, exponential, asymmetric) across risk-stratified simulation environments
• Key finding: PPO clip and long rollouts are interdependent stabilizers — neither alone sufficient for long-horizon policy stability; produced a research-quality ablation study with quantitative comparisons across all algorithm and reward variants

View on GitHub →

Jan 2026

Multi-Sensor Fusion Pipeline for Autonomous Vehicle Perception

Built a Camera/LiDAR/GPS sensor fusion pipeline on the KITTI dataset — synchronizing multimodal streams at 10Hz (3ms jitter), implementing calibrated projection with 85% visibility and <5px reprojection error, and automating quality gates across temporal, spatial, and radiometric dimensions. Architected for extensibility toward downstream object detection and tracking.

View on GitHub →

Dec 2025

2D Robot Navigation Simulator with LiDAR-Based Obstacle Avoidance

Built a Python navigation simulator integrating RRT path planning with a hybrid adaptive multi-gain P-controller and 360° LiDAR perception for real-time collision avoidance, achieving 90%+ goal-reaching success in obstacle-dense environments.

View on GitHub →

Nov 2025

Real-Time Kinematics and Trajectory Control for UR5 6-DOF Robot

• Replaced an unreliable iterative IK solver with a differential velocity-level controller built from scratch, achieving sub-3mm trajectory accuracy across circular, square, and sine wave end-effector paths
• Developed a React-based interactive visualization dashboard for real-time monitoring of joint states, velocities, and end-effector pose during trajectory execution

View on GitHub →

Nov 2025

Motion Planning for Ackermann Vehicles using Dubins-Constrained RRT* Variants

• Benchmarked three Dubins-constrained planners (RRT, Bi-RRT, P-RRT*) across 300 trials in diverse obstacle environments for nonholonomic Ackermann-steered vehicles
• Identified BiRRTStarDubins as optimal with 10× faster execution (0.043s) and 100% success rate; quantified 292% computational overhead of nonholonomic constraints

View on GitHub →

Jul 2021 – May 2024

Lead Software Engineer / Team Lead

Giottus Technologies Pvt Ltd., Chennai, India

Distributed Systems · Platform Architecture · Engineering Leadership

Leadership & Org Impact

• Joined as an early engineer on a sub-10 team; grew into technical lead owning end-to-end platform delivery as the company scaled to 1M+ customers — drove hiring, interview rubrics, and onboarding playbooks
• Led Agile delivery across a 10+ person team — mentored engineers on system design and clean architecture, pioneered Git Flow and CI/CD practices org-wide, accelerating release cadence by 40% and reducing defects by 25%

Systems Design & Scale

• Designed and owned production-scale distributed systems using Java, Spring Boot, Kafka, Docker, and Kubernetes — achieving 300% throughput improvement and 45% downtime reduction under 10× traffic spikes
• Led monolith-to-microservices migration — decomposed a tightly coupled codebase into independently deployable services; integrated Prometheus/Grafana observability achieving 99.9% uptime and 30% faster incident detection

Implementation & Execution

• Built gRPC/WebSocket communication layers and REST APIs with full lifecycle management; implemented JWT-based auth and RBAC from scratch; integrated Stripe, KYC providers, and exchange APIs with key lifecycle management
• Drove TDD culture with JUnit scenario testing, parallel wallet simulation tests, and A/B rollouts; diagnosed a race condition in live wallet updates via transaction replay and resolved with row-level locking

Dec 2018 – Nov 2021

Machine Learning Engineer

CloudFabrix Software Pvt Ltd., Hyderabad, India

ML Pipelines · AIOps · Experimentation Platform

ML Platform & Tooling

• Built a core internal ML experimentation platform from scratch — evolving a plugin-based Python SDK into a full service supporting experiment tracking, model versioning, and reusable supervised/unsupervised pipelines adopted org-wide; reduced model-to-production time by 40%
• Collaborated with domain experts through human-in-the-loop validation workflows to diagnose model failure modes and refine behavior before production deployment

Pipelines & Anomaly Detection

• Designed real-time data ingestion pipelines with automated schema validation and quality checks over high-volume telemetry, reducing latency by 20% and enabling sub-second anomaly detection
• Applied unsupervised clustering (KMeans, DBSCAN, HDBSCAN) over large-scale infrastructure telemetry, achieving ~90% false-alert suppression while preserving critical failure signals

Forecasting & Monitoring

• Deployed Prophet and ARIMA-based time-series forecasting on rolling 90-day sensor windows, using forecast-residual bands to detect abnormal behavior — achieving 85% prediction accuracy and reducing unplanned downtime by 30%
• Designed data storage schemas and partitioning strategies for large-scale time-series and event data to support analytics and ML workflows

Nov 2017 – Nov 2018

Software Developer

Geazy Technologies LLP, Chennai, India

• Re-engineered core ad posting platform components, improving performance by 15% and reducing operational costs by 10%
• Built efficient data transformation parsers in PHP for large-scale data processing pipelines
• Resolved production issues in chatbot integrations, reducing downtime by 40%