Kunal Mahato
Open to work
Samsung  ·  Turvo  ·  Virginia Tech

Engineer who ships, now with AI in the loop.

San Francisco Bay Area
Pitched for →

AI Engineer building agentic RAG pipelines on Gemini, OpenAI, LLaMA, with 5+ yrs shipping production code at Samsung and Turvo before grad school.

Based
San Francisco Bay Area · open to relocation
Availability
Open to full-time roles
Now
Benchmarking frontier LLMs on sensitive data
Kunal Mahato
Kunal Mahato
01Impact

By the numbers

1.48B+
Devices reached at Samsung
3 yrs on Android SystemUI: Quick Panel, Notifications, Status Bar
216K+
Records migrated
Zero data loss · MySQL → MongoDB · Turvo Intermodal launch
99.9%
Uptime
UPass Manager · 10K+ student records on Kubernetes
50%
API speedup
Optimized GraphQL/REST endpoints on UPass Manager
80%
F1 classification
Agentic RAG pipeline at VT research lab
3.93
GPA at Virginia Tech
M.Eng. Computer Science, May 2025
02For the ai engineer

Why you should hire me as AI Engineer

Pitched for →

I'm not an AI person who learned to code. I'm an engineer with 5+ yrs in production (Samsung SystemUI, Turvo platform) who pivoted into AI when it became clear it would reshape how software gets built. That means I ship LLM features that actually work in production, with evals, guardrails, and the plumbing around the model, not just the prompt.

Proof
Agentic RAG pipeline: 80% F1 on sensitive-data evalRead the full story →DoDEx: hackathon RAG for Defense Contract NoticesRead the full story →
03Experience

Where I've shipped

Jul 2025 – Apr 2026
Remote

Research Assistant at Virginia Tech

Show stack
AIRAGPythonPyTorchLLMs
  • Built data-processing pipelines for LLMs in Python integrating OpenAI, Llama, and Gemini APIs through implementation of data-selection strategies including seed-based, embedding-based, alongside domain expert in the loop.
  • Designed an agentic RAG pipeline (PyTorch + SentenceTransformers) with tool-use for top-K selection, improving relevance and recall and achieving ~80% F1 in classification tasks.
  • Automated reporting with Pandas/NumPy, producing results across 12 model setups and reducing run time by 3x.
May 2021 – Jun 2023
Hyderabad, India

Software Engineer at Turvo Inc.

Show stack
Full-stackPythonNodeElasticsearchETLCI/CD
  • Enhanced global search and filter efficiency by 10% through advanced Elasticsearch indexing and GraphQL API implementation, improving data retrieval speed and accuracy for supply-chain operations.
  • Architected a resilient Python-based ETL pipeline to launch Intermodal shipment mode, migrating 216,000+ multi-leg records between heterogeneous data stores (MySQL to MongoDB) with zero data loss during a critical schema redesign.
  • Managed reorganization of the Rules Framework into a dynamic Business Rule Management System, enforcing complex role-mode-status compliance constraints across 10+ enterprise clients while decoupling business logic from the core codebase.
  • Built Jenkins CI/CD pipelines for isolated dev setups, cutting deployment time by 40% and improving release reliability.
  • Implemented a resilient Python library for HashiCorp Vault, ensuring secure secret injection with 99.99% uptime for 10 clients.
Dec 2019 – Dec 2019
Suwon, South Korea

R&D Engineer at Samsung HQ

Show stack
AndroidJavaFramework
  • Integrated backend UI components in Java using Android Framework with system services for foldable-specific notifications, contributing to the global launch of the world's first foldable phone at Samsung HQ.
Jun 2018 – Apr 2021
Noida, India

R&D Engineer at Samsung

Show stack
AndroidJavaSystemUIPerformance
  • Maintained and optimized full-stack codebase for 5 OS iterations, enhancing performance for 1.48+ billion devices.
  • Optimized System UI (Quick Panel & Notifications) in Java for low end devices, cutting memory use 20% and battery drain 20%, reducing crash rate 85% and boosting user satisfaction to 4.4/5 across 1M+ devices.
  • Handled 5 Application and Framework modules - Settings, Calendar, QuickPanel, Status-Bar and warning notifications.
  • Upgraded the customer satisfaction metrics by 10% through implementation of Search feature for Quick Panel.
  • Collaborated cross-functionally with teams to build backend data-porting solutions integrating 20+ network operators.
04Academic record

Education

Virginia Tech

Aug 2023 – May 2025 · Alexandria, VA
M.Eng. Computer Science & Applications
GPA 3.93 / 4
Coursework
Software Engineering · Machine Learning · AI tools for SW Delivery · Cloud Computing · Usability Engineering · Intro to AI · Web App Development · Urban Computing

NIT Allahabad

Jul 2014 – May 2018 · Prayagraj, India
B.Tech. Computer Science & Engineering
GPA 3.53 / 4
Coursework
Data Structures · Algorithms · DBMS · Distributed Systems · Operating Systems · Networks · Cryptography · OOM
05Key stories

Stories, for your next screen

Each story uses Situation · Task · Action · Result. The Ask about Kunal chat can expand any of these on the fly.

Shipping the world's first foldable at Samsung HQ

Situation
In 2019, Samsung was racing to ship the world's first foldable phone. I'd been on Android SystemUI in India for a year.
Task
Ship foldable-specific notifications, a brand new interaction surface, in time for global launch out of Samsung HQ in Seoul.
Action
Moved to Seoul HQ for 3 weeks. Integrated backend UI components in Java with Android Framework system services, handling the new folded/unfolded states across the notification pipeline. Pair-coded with Korean teammates in a time-pressured environment.
Result
Shipped in time for the Galaxy Fold global launch. Code ran on every foldable shipped that year and is the foundation of the modern foldable notification stack.
06Selected work

Projects

6 projects
Jan 2025 – May 2025

UPass Manager

Virginia Tech
UPass Manager screenshot
Full-stack

Kubernetes-managed Next.js web app for 10K+ student records with 99.9% uptime.

  • Architected Kubernetes-managed Next.js, React, Node.js web app ensuring 99.9% uptime for 10K+ student records.
  • Cut manual entry 95% with AWS Lambda automating S3 → RDS ingestion, boosted pickups via SES notifications.
  • Improved API speed 50% via optimized GraphQL/REST endpoints (Express, Apollo, JWT).
  • Halved student wait times with a 100ms NFC validation pipeline using Socket.IO.
Show stack
KubernetesAWSReactNode.jsMySQLNFCChart.js
Apr 2025

LifexAI

University of Maryland · Bitcamp
LifexAI screenshot
AI

AI detective web app: image understanding + transactional data for anomaly detection.

  • Integrated Google Gemini Vision API, Capital One Nessie API, and real-time Chart.js dashboards, boosting anomaly detection accuracy 30%.
  • Accelerated data loading 50% with React Query caching + Next.js APIs on Cloudflare Workers.
  • 99.9% uptime via Vercel + automated CI/CD with Docker.
Show stack
Gemini VisionCapital One NessieVercelMongoDBTailwind
Aug 2024 – Dec 2024

Gentrification Prediction

Virginia Tech
Gentrification Prediction screenshot
ML / Data

ML pipeline fusing Census, Zillow, Eviction Lab data to predict gentrification risk.

  • Fused US Census, Zillow ZHVI, Eviction Lab data into unified time-series geospatial DB via ETL.
  • Built binary classifiers (LR, RF, LightGBM) with 5-fold CV + GridSearchCV, achieving 0.77 ROC-AUC.
  • Used SHAP to interpret PCA-reduced features, mapping back to socioeconomic variables.
  • Built SARIMAX models to forecast 5-year SEIFA scores, flagging at-risk counties.
Show stack
Feature EngineeringGeospatialMLSHAPSARIMAX
Jan 2024 – May 2024

Book Wonders

Virginia Tech
Book Wonders screenshot
Full-stack

E-commerce platform: 1000+ transactions, JWT auth, 15+ REST APIs.

  • Scalable e-commerce app with login, browsing, cart, and checkout, supporting 1000+ transactions.
  • Responsive pages with React + Context API + Redux, improving load and data flow up to 30%.
  • 15+ RESTful APIs in Java with MySQL integration.
  • Deployed on AWS with JWT auth for secure sessions.
Show stack
ReactReduxJavaMySQLAWS
Apr 2024

DoDEx

Bitcamp Hackathon
DoDEx screenshot
AI

AI ingestion pipeline for Defense Contract Notices, 30% more accurate than regex.

  • Python + RAG pipeline parsing unstructured Defense Contract Notices, 30% higher accuracy than regex.
  • Built a ground-truth validation framework to benchmark LLMs and prevent hallucinations in mission-critical reporting.
Show stack
LLMRAGElasticsearchWeb Scraping
Aug 2023 – Dec 2023

Review Me

Virginia Tech
Review Me screenshot
Full-stack

Peer review platform on MERN: hashtag search, upvotes, 95% test coverage.

  • Full-stack peer review platform on MERN with MVC architecture.
  • Hashtag search, notifications, upvote/downvote, comments for content discovery.
  • 95% test coverage with Mocha + Chai: unit and acceptance.
  • Improved UX 20% with post editing, search, notifications, guest functionality.
  • Secure auth via JWT with encrypted stateless communication.
Show stack
MERNJWTMochaMVC
End of portfolio · begin conversation

Let's talk.

The fastest way to a reply is email. If you just want to browse, grab the PDF.

mahatokunal41@gmail.com Resume LinkedIn GitHub
Kunal Mahato · Portfolio · 2026Press ⌘K / Ctrl+K to navigate