Parikshit Kumar | ML Engineer

BACKGROUND

About

I build machine learning systems that work in the real world—not just benchmarks. My focus spans real-time computer vision, generative modeling, and robust data infrastructure, taking complex requirements from architecture through to production at scale.

I lead engineering at the intersection of research and product: driving system resilience, optimizing for latency, keeping metrics observable. I translate ambiguous problems into structured technical plans and deliver systems that move the needle on core business outcomes.

WORK HISTORY

Experience

Oct 2025 — Present

Chief AI Engineer

Plexor

Own the architecture and deployment of real-time object detection and vision-language models for retail surveillance.
Built fault-tolerant edge AI pipelines that hold under constrained, adversarial store conditions.
Established new detection precision benchmarks for the platform's core vision capability.

Jul 2024 — May 2025

Junior Specialist

University of California, Riverside

Deployed deep learning classifiers for mosquito orientation analysis—96% accuracy.
Built end-to-end image processing pipelines using thresholding and edge detection to isolate critical biological features.
Enabled interdisciplinary teams to scale feature extraction workflows across large research datasets.

Jan 2023 — Jun 2023

Research Engineer

University of Guelph

Fused image segmentation with LLM-driven analysis to assess soil properties—18% accuracy gain over prior methods.
Cut data processing time by 40% through automated tooling; measurably improved indicator precision across the board.
Collaborated directly with domain scientists to translate model outputs into actionable agricultural insights.

Mar 2022 — Jul 2022

Computer Vision Researcher

Avawatz · UT Dallas · IIT Bombay

Built object detection systems for autonomous off-road navigation using Faster R-CNN and YOLO—15% accuracy improvement.
Leveraged 3D simulation environments to achieve a 20% gain in segmentation precision.
Cut large-dataset training time by 25% through preprocessing pipeline optimization.

Jan 2021 — Sep 2022

Machine Learning Engineer

Lokavidya · IIT Bombay

Architected NLP-driven knowledge curation engines that drove measurable improvements in search relevance and retrieval speed at scale.
Structured chaotic raw datasets through classification and clustering, eliminating redundancy across the board.
Automated anomaly detection pipelines, cutting manual verification overhead by 20%.

EXPERTISE

Skills

Core Capabilities

Deep Learning

NLP

LLMs

Computer Vision

Generative Modeling

Diffusion Models

Machine Learning

Reinforcement Learning

Data Analysis

Data Visualization

Statistics

Graph Neural Networks

Languages & Frameworks

Python

C++

C

Keras

TensorFlow

PyTorch

PyTorch Lightning

FastAI

OpenCV

Numpy

Matplotlib

Pandas

Hugging Face

SQL

Flask

BeautifulSoup

XGBoost

Tableau

MLOps

GCP

AWS

Kubernetes

Docker

Git CLI

Jupyter Notebooks

Anaconda

Linux Terminal

ROS

CUDA

MPS (Apple)

Apache Spark

Tensorboard

Firebase

MS Azure

MATLAB

SELECTED WORK

Projects

ApollodB: Psychoacoustic EQ Generation System

SOTA Project · 2024 to Present

Designed a TensorFlow-based psychoacoustic EQ engine deployed on Google Cloud—state-of-the-art results on the DEAM dataset.
Built custom neural architectures generating highly personalized audio profiles grounded in perceptual science.
Open-sourced the full platform, including the interactive validation suite, for community adoption.

PythonTensorFlowGoogle CloudPsychoacousticsDEAM

Project page Code on GitHub

Generative Networks and VAEs to Assist Artists

SRM IST · Jan 2023 to Jun 2023

Built Variational Autoencoders for sketch refinement—demonstrated 30% improvement in artist workflow efficiency.
Achieved 90.6% accuracy on the CUHK Face Sketch Database; presented findings at ICCET 2023.
Designed the architecture to preserve high-frequency detail, closing the gap between generative output and artistic intent.

PythonKerasTensorFlowOpenCVVAE

Read paper

Energy Price Forecasting Using LSTMs and ARIMA

SATU · 2022

Built hybrid LSTM/ARIMA forecasting pipelines—18% accuracy lift on complex Australian energy market data.
Integrated NLP-driven trend analysis to surface broader market context for key stakeholders.
Presented methodology and scalability analysis at the AIJR Conference 2022.

PythonLSTMARIMANLP

Read paper

PersonaForge: Fine-Tuned LLM Assistant

UC Riverside · Dec 2024

Fine-tuned a personal LLM using llama.cpp for contextual, day-to-day task automation.
Engineered for local-only deployment: strict data privacy with zero latency sacrifice.

PythonLLMllama.cppPersonal AssistantNLP

FormScribe: AI-Powered Web Form Automation

Apr 2025

Automated complex data entry by bridging Selenium with a local LLaMA 2 7B model—fully offline, zero data exposure.
Architected for robustness and data security across diverse, real-world browser environments.

PythonSeleniumOllamaLLaMA 2 7BAI Automation

Code on GitHub

ACADEMIC BACKGROUND

Education

Sep 2023 — Dec 2024

Master of Science, Computer Science