Sep 2021 – May 2025
B.Tech, AI & Machine Learning
D.J. Sanghvi College of Engineering · Mumbai
Machine learning, deep learning, computer vision, NLP, and data structures & algorithms.
- Mumbai University
- AI & ML
Kunj Pandya
Software Engineer · Distributed Systems & Applied ML
📍 Hoboken, NJ · Open to relocation
MS CS @ Stevens, May 2027 · Open to new-grad SWE and ML roles
I build distributed systems from their primitives and ML systems that ship: a hand-built Chord DHT with gRPC event streaming, production NLP at 10K+ requests a day, and retrieval on Cloudflare's edge.
10K+
Requests/Day in Production
1st
QuackHacks '26
Top 100
HackHouse SF
Journey
From Mumbai to Hoboken: every role, paper, project, and win on one line, in order.
21 entries · Sep 2021 → Jun 2027 · 5 running now
Entries are evenly spaced, so busy years take more room. Hover or tap one to see how long it ran.
About Me
I'm a software engineer finishing an MS in Computer Science at Stevens Institute of Technology (May 2027), after a B.Tech in Artificial Intelligence and Machine Learning from D.J. Sanghvi College of Engineering in Mumbai. I'm looking for new-grad software and ML engineering roles starting in mid-2027.
I like building systems from their primitives. I implemented a Chord DHT by hand and layered gRPC event streaming on top of it, built a task orchestrator that holds up against starvation and retry storms under burst load, and designed a document pipeline where every generated claim has to resolve to its source or the build fails.
Today I build AI-assistant features for Fini's iOS app in Swift and SwiftUI, shipping through senior-reviewed pull requests. Before that I owned the backend of a placement platform at D.J. Sanghvi that served 500+ students at 10K+ requests a day, and I co-authored a peer-reviewed paper on AI ingredient recognition.
Highlights
1st Place, QuackHacks '26
Won the 24-hour hackathon with I Wizard, a real-time 3D browser game built with React Three Fiber.
2026
Top 100, HackHouse SF
Selected for the San Francisco hacker house, working alongside founders from SuperMemory, AirGarage, Omi, and Browser-Use. Built FleetIQ there and pitched it to AirGarage.
Dec 2025
Peer-Reviewed Publication
Co-authored "Revolutionizing Culinary Experiences: AI-Driven Ingredient Recognition and Personalized Recipe Recommendation System" in Library Progress International, and first-authored its system architecture section.
2024
Published Chrome Extension
QuickFill, a keyboard-driven extension for snippets, autofill, and templates, is on the Chrome Web Store with 19+ active users.
2026
OCI 2025 Certified AI Foundations Associate
Oracle Cloud Infrastructure certification. Also completed Java Essential Training: Objects and APIs, and Claude Code in Action.
2025
Education
Master of Science
Computer Science
Stevens Institute of Technology
Hoboken, NJ
Sept 2025 – May 2027
Key Coursework:
Distributed SystemsOperating SystemsConcurrent Programming+4 more
Bachelor of Technology
Artificial Intelligence and Machine Learning
D.J. Sanghvi College of Engineering, Mumbai University
Mumbai, India
Sept 2021 – May 2025
Key Coursework:
Machine LearningDeep LearningComputer Vision+2 more
Experience
Software Engineer Intern, iOS (AI Features)
CurrentFini
Remote
May 2026 – Present
- ▸Build AI-assistant features in Swift and SwiftUI (MVVM, Swift Package Manager) that turn natural-language intent into structured actions for task assignment and routine building.
- ▸Migrate legacy UIKit screens to SwiftUI behind feature flags, so rebuilt screens roll out incrementally without breaking what already ships.
- ▸Ship through senior-reviewed pull requests in a multi-contributor codebase (~15 active contributors), with SwiftLint-enforced style and localization across English, German, and Spanish.
SwiftSwiftUIUIKitMVVMSwift Package ManagerFeature FlagsSwiftLint
Graduate Research Assistant, Distributed Systems
CurrentStevens Institute of Technology
Hoboken, NJ
Nov 2025 – Present
- ▸Research energy efficiency in heterogeneous federated learning (HeteroFL), focusing on where cold-start training wastes energy.
- ▸Built a distributed profiling pipeline with TensorFlow.js, Socket.IO, and Redis that collects per-layer power metrics in real time from 100+ heterogeneous edge devices.
- ▸Traced cold-start energy overhead to specific bottleneck layers; the resulting optimizations cut training energy 35% with under 2% accuracy loss. Findings submitted for publication.
TensorFlow.jsSocket.IORedisFederated LearningDistributed SystemsEdge ComputingProfiling
Research Intern – Placement Intelligence Platform
D.J. Sanghvi College of Engineering
Mumbai, India
May 2024 – June 2025
- ▸Owned the backend for a placement platform serving 500+ students, with REST APIs handling 10K+ requests a day at under 200 ms p95 latency during concurrent resume uploads.
- ▸Trained a custom spaCy NER resume parser on 2,000 labeled resumes (70/15/15 split), reaching 95% F1 against a 78% regex baseline, and served it through FastAPI.
- ▸Fine-tuned MobileNetV2 for real-time interview-confidence detection at 30 FPS and 88% accuracy, and cut inference latency 40% with quantization.
Node.jsFastAPIREST APIsspaCyNERTensorFlowMobileNetV2Microservices
Undergraduate Research Assistant
D.J. Sanghvi College of Engineering
Mumbai, India
June 2022 – May 2024
- ▸Co-authored a peer-reviewed paper in Library Progress International (2024) on AI ingredient recognition and personalized recipe recommendation, and first-authored its system architecture section.
- ▸Built the ingredient recognition model on MobileNetV2 and paired it with a hybrid recommender combining collaborative filtering and TF-IDF content matching over an 8,269-recipe catalog.
TensorFlowMobileNetV2Computer VisionTF-IDFRecommendation SystemsFlutter
Featured Projects
A selection of projects demonstrating expertise in distributed systems, AI/ML, and scalable backend architecture
AI Systems & Data Pipelines
Edara — Procedure Intelligence Engine
May 2026 – Present
Turns technical PDFs like install guides, service manuals, and statements of work into structured, source-grounded task data, with the LLM confined to meaning and every claim traced back to its source.
Impact: A provenance resolver maps every generated claim to an exact source span, and unresolvable provenance fails the build. 13 test modules, including human-read golden ground truth, run against a 13-document corpus of networking, appliance, and statement-of-work documents.
PythonFastAPISupabasePostgreSQLRedis+3 more
Private repository
Technical Skills
Languages
PythonJavaTypeScript / JavaScriptGoC / C++SwiftSQLBash
Distributed Systems & Backend
Chord DHT & Consistent HashinggRPC / ProtobufFastAPINode.js / ExpressRedis (Queues, Pub/Sub, Caching)PostgreSQLWebSocketsFault Tolerance & IdempotencyConcurrency
AI / ML
PyTorchTensorFlowComputer Vision (YOLOv8, ResNet-50, MobileNetV2)NLP (spaCy NER)RAG & Vector SearchLLM Integration (Tool Calling, Streaming)XGBoostscikit-learn
Cloud & Infrastructure
Docker / ComposeAWS (EC2, S3)Cloudflare Workers & Durable ObjectsSupabaseVercelGitHub Actions CI/CDLinux
Mobile & Frontend
SwiftUI / UIKitReactNext.jsTailwind CSSReact Three FiberUnity (C#)
Engineering Practice
Git & PR Review WorkflowFeature Flagspytest / JUnitMavenProfiling & BenchmarkingAgile
Always learning and exploring new technologies to build better systems
Get in Touch
I'm always open to discussing new opportunities, collaborations, or just having a chat about distributed systems and AI/ML.
Location
Hoboken, NJ · Open to relocation
© 2026 Kunj Pandya. Built with Next.js, TypeScript, and Tailwind CSS.