Jignasu Pathak

Degree: Masters of Computer Science from Virginia Tech(2023 grad) Experience: 2 Years of SDE experience at Amazon

I am a Software engineer with 2 years of experience building and maintaining highly scaled trustworthy computer vision system in production.

My current work centers on the Berkley Single Cell Computational Microscopy(BSCCM) benchmark pipeline for image to image translation, predicting whiteblood cell flourescence from unlabelled cell images. This project is a proof of concept for my approach to building trustworthy machine learning systems: start with a clear, reviewable baseline and iterate with disciplined comparison instead of chasing complexity from the start.

View Projects Open the resume

Professional snapshot

  • Experience As an SDE at Amazon I developed and maintained business critical(Tier-1) highly scaled systems for automated defect detection using computer-vision and customer facing full-stack application. My Masters Thesis was studying how AI explanation can affect accuracy of Human-AI team and how human tendancy to avoid cognitive load makes us prone to overreliance, shallow judgment, and uncritical acceptance of AI outputs. I want to combine these two experience and apply to the domains(eg. computational microscopy) where silent model failures are expensive and requires scalable verification with human-centered design.
  • Skills Cloud: AWS (Certified), Google Cloud Languages: Python, Java, TypeScript Databases: MySQL, DynamoDB Frameworks: PyTorch, TensorFlow, Keras, JAX, React, Flask Tools: Git, CloudWatch, Docker, TensorBoard Engineering Practices: CI/CD, Integration Testing, Rollback Readiness, Observability, Agile Delivery Concepts: Machine Learning, Computer Vision, Reproducibility, Evaluation Design

Professional background and resume

Review prior experience here

Open resume page Download PDF

Projects

Selected projects that reflect the problems I tackled, the technical decisions I made, and the outcomes they produced.

See all projects

Project Computational Microscopy

Building the first comparison-ready baseline Image-to-image dense prediction

BSCCM microscopy image-to-image pipeline

A reproducible image-to-image benchmark pipeline built on the BSCCM dataset. The project uses 23 label-free microscopy views to predict 6 fluorescence channels, with an emphasis on controlled data handling, explicit run configuration, and stable evaluation before broader model comparisons.

Why it matters Microscopy model comparisons are easy to distort when data handling, preprocessing, and evaluation are inconsistent. This project focuses on building a baseline that is stable enough to trust before adding more model complexity.

Current checkpoint Locking down a baseline that is actually trustworthy to compare: fixed data handling, explicit run configuration, stable metrics, and saved artifacts that make errors easier to inspect.

Microscopy Reproducibility Evaluation

Writings

Technical notes covering project scope, experimental structure, implementation choices, and results.

Browse all writing

Case study

Methods

Tools

Reusable tooling built to support project with diagrams, workflows, dashboards, and internal utilities that strengthen future projects.

Open tools library

Tool Tooling + diagrams

Actively used Diagram editor

Neural Architecture Diagram Editor

An interactive diagram editor for composing clean neural-network architecture visuals while learning, designing, and documenting models. Built to make architecture reasoning inspectable and reusable instead of redrawn by hand.

Tooling Architecture Diagrams