Vellore Institute of Technology, Chennai

DataScience&AppliedML

Software Systems Engineer
“From raw data...”

“I build software systems and back them with applied machine learning. My work usually starts with a real problem and goes from raw data to a model to something deployed that people can actually use.”

“What drives me: building for people who are underserved — assistive tech, disaster response, clinical screening — and taking an idea from raw data to something deployed and working.”

Available • Applied ML & Systems
Harshit Singh — Professional
Tap or hover to flip
92.4% AccuracyReal-Time SPARC Sign Recognition (12K Samples)
98.4% Val AccCoastal LSTM Flood & Microclimate Forecast
>95% UptimeTRIDENT Distributed Telemetry & Risk Routing

Featured Demos

Interactive previews of predictive models & edge perception loops

TRIDENT — Predictive AI & Command Center

Live Telemetry & 12-Node Risk Routing

Stacked PyTorch LSTM microclimate forecasting, NetworkX graph diffusion risk routing, and real-time P1-P5 victim triage telemetry under sub-50ms latency.

SPARC — Sign Language Recognition

Multimodal 3D Mesh & Voice Synthesis

92.4% accuracy real-time gesture recognition pipeline supporting ISL and ASL with multimodal fusion — hand landmarks, facial emotion, and voice synthesis.

STEMBOT — Indoor Autonomous Mobile Robot Console

STATION ALPHA // DIGITAL TWIN
ROS 2 Jazzy • RViz 2 & Gazebo

ROS 2 Jazzy — Digital Twin Simulation

LiDAR Perception Loop • 3,600 pts/sec stream

High-fidelity Gazebo simulation with RViz 2 costmap inflation, 2D LiDAR /scan filtering, and dynamic Nav2 obstacle traversal with 95% collision-free passing.

STATION BETA // PHYSICAL VALIDATION
On-Chassis Sensor Fusion

Indoor Autonomous Mobile Robot — Physical Chassis

15 Hz EKF • 60% Spatial Drift Cut & 69.2% Angular Error Reduction

Live corridor testing validating extended Kalman filter (EKF) state estimation fusing wheel encoders and 6-DOF IMU under real-world kinematic floor slip.

Project & Competition Gallery

Live captures from robotics testing, national hackathons, and edge ML deployments

SPARC Embedded System
Edge Vision & MLSPARC Embedded System
TRIDENT Command Dashboard
Predictive AI SystemsTRIDENT Command Dashboard
Terrain Scout Rover
Autonomous RoboticsTerrain Scout Rover
DrishtiGuide Components
Assistive ML WearableDrishtiGuide Components
RFID Scanner Telemetry
Sensor IntegrationRFID Scanner Telemetry
Aerial Perception Drone
Autonomous SystemsAerial Perception Drone
Telemetry Bench Testing
Signal ValidationTelemetry Bench Testing
SPARC Embedded System
Edge Vision & MLSPARC Embedded System
TRIDENT Command Dashboard
Predictive AI SystemsTRIDENT Command Dashboard
Terrain Scout Rover
Autonomous RoboticsTerrain Scout Rover
DrishtiGuide Components
Assistive ML WearableDrishtiGuide Components
RFID Scanner Telemetry
Sensor IntegrationRFID Scanner Telemetry
Aerial Perception Drone
Autonomous SystemsAerial Perception Drone
Telemetry Bench Testing
Signal ValidationTelemetry Bench Testing
SIH Finals — SPARC Showcase
Smart India HackathonSIH Finals — SPARC Showcase
SIH Grand Team Photo
National FinalistSIH Grand Team Photo
Technovation Trophy
1st Prize WinnerTechnovation Trophy
Innovate X Impact
Winner — AI CategoryInnovate X Impact
Bureau of Indian Standards
National Project ExpoBureau of Indian Standards
Engineering Honors
Competition RecordEngineering Honors
Namo Nirvana Team
Technical LeadershipNamo Nirvana Team
SIH Finals — SPARC Showcase
Smart India HackathonSIH Finals — SPARC Showcase
SIH Grand Team Photo
National FinalistSIH Grand Team Photo
Technovation Trophy
1st Prize WinnerTechnovation Trophy
Innovate X Impact
Winner — AI CategoryInnovate X Impact
Bureau of Indian Standards
National Project ExpoBureau of Indian Standards
Engineering Honors
Competition RecordEngineering Honors
Namo Nirvana Team
Technical LeadershipNamo Nirvana Team

Core Engineering Pillars

The three foundational disciplines backing my systems

Data Science & Predictive Modeling

Autoregressive multivariate LSTMs, directed graph risk diffusion (NetworkX), Scikit-Learn classification pipelines, and multi-criteria decision analysis (MCDA) for triage.

Applied ML & Computer Vision

Real-time gesture pipelines (92.4% accuracy), MediaPipe 468 landmark face meshes, in-browser WebGL inference, and high-frequency LiDAR perception streams.

Software & Data Systems Engineering

Full-stack Next.js/React applications, Node.js RESTful APIs (70+ endpoints), shared-memory optimizations (/dev/shm), and Dockerized backend microservices.

Technical Arsenal

Verified technical stack organized across four engineering lifecycle stages

//Recurrent architectures, stochastic state estimation, and statistical modeling
PythonCore Language
PyTorchDeep Learning
Scikit-LearnStatistical ML
N
NumPyVectorized Math
PandasData Ingestion
Stacked LSTMsTime-Series
Extended Kalman FilterState Estimation
NetworkXGraph Diffusion

System Diagnostics

Style 24 IDE Code Stream & interactive telemetry shell

15 Hz SYNC
01
import torch, cv2, ros2_telemetry, ekf_filter
02
async def verify_applied_ml_pipeline():
03
buffer = await shm.open("/dev/shm/telemetry", rate="15Hz")
04
ekf.validate(stream=raw_encoder_logs, trials=20)
05
lidar.filter_pointcloud(rate=3600_pts_sec, max_latency=50)
06
sparc_model.eval(dataset=12000_samples, mode='multimodal')
07
return {"status": 200, "pipeline": "READY_FOR_DEPLOYMENT"}

Start a Conversation

Drop an email or connect through professional networks

Connection Nodes

Whether you want to discuss predictive ML models, time-series forecasting, data engineering pipelines, or software systems architecture, reach out anytime.

Transmission Portal