Hi, I'm Sneha Grian Joshua
|
I am a software engineer with 3+ years of experience and a Master's in Information Technology from Arizona State University. Most of my work has been full stack: I have built and shipped complete websites and applications end to end, both inside a company and independently as a freelancer, owning everything from the database schema and the API layer through to the interface people actually click on.The core of that work is backend engineering. I have built and maintained REST APIs and microservices across 10+ production applications using Node.js, Spring Boot, and FastAPI, backed by MySQL and PostgreSQL. A lot of it has been ERP and operations software, covering inventory and orders, supplier and purchase workflows, user management and role-based access, dashboard reporting, and payment and transaction flows. I have cut p95 API latency by 45% by profiling slow endpoints and fixing the queries underneath them, raised throughput by 38% by rewriting blocking logic as asynchronous workflows, and reduced recurring production incidents by 40% through better testing, validation, and CI/CD checks.On the front end I work in React, Next.js, and Angular, and I enjoy the craft side of it more than most backend engineers do: hand-written CSS with proper design tokens, responsive layouts with fluid typography, charts drawn in pure CSS rather than pulled from a library, and animation tuned so it does not cost the browser more than it needs to.More recently I have moved into AI and RAG systems, and it is where most of my curiosity currently sits. I am a Databricks Certified Generative AI Engineer, and I have built agentic workflows with LangGraph where the agent plans its own sub-questions, grades whether the evidence it retrieved is good enough, loops back to search again when it is not, then verifies its own answer against the source before showing it. I work with vector search across Chroma and Pinecone, embedding and chunking strategies, tool calling, and the practical side of shipping these systems, like caching answers so the model runs as little as possible.I am currently based between Florida and Arizona and open to relocation. If you are building something interesting, I would like to hear about it.
Backend & APIs
Node.js and Spring Boot microservices, REST API design, and SQL schema and query work across 10+ production applications
Frontend & UI
React, Next.js, and Angular interfaces with hand-written CSS, design tokens, responsive fluid layouts, and dependency-free charts
AI & RAG Engineering
LangChain and LangGraph agents, RAG pipelines, vector search, and tool calling, backed by Databricks Generative AI certification
Performance & Reliability
Query profiling, indexing, and async rewrites that cut p95 latency by 45% and recurring production incidents by 40%
Work Experience
Application Developer
- ●Built a Python and FastAPI AI chatbot backend with REST API endpoints handling user input validation, structured prompt formatting, LLM API integration, and guided response generation
- ●Integrated wearable ring, mood-tracking, and research-notes data into a RAG-style workflow, converting activity, sleep, mood, and HR/HRV-style signals into JSON context for personalized AI responses
Software Developer
- ●Built and maintained full-stack features for 10+ client applications using Node.js microservices, REST APIs, SQL databases, and React, owning backend API development, frontend integration, testing, and production support
- ●Built secure backend workflows for a medical inventory ERP-style dashboard, including JWT-based authentication, OAuth login, protected API routes, role-based access control, and middleware-level authorization
- ●Built reusable API modules covering user management, inventory and orders, supplier and purchase workflows, dashboard reporting, file import and export, and notification triggers, so the same code supported multiple project areas
- ●Reduced p95 API latency by 45% by profiling slow endpoints, reviewing SQL execution plans, finding queries doing repeated full table scans, adding composite indexes, and removing redundant database calls on high-traffic routes
- ●Increased backend throughput by 38% by rewriting blocking service logic as asynchronous Node.js workflows, cutting sequential database operations, improving Promise-based execution, and tightening API payloads
- ●Cut recurring production incidents by 40% by debugging Linux environments, tracing failed requests, and fixing validation and authentication edge cases, adding Jest unit tests, REST API integration tests, and CI/CD checks before deployment
Research Intern
- ●Developed Spring Boot microservices and REST APIs within a distributed backend architecture, implementing controller-service-repository layers, request validation, exception handling, and service-level business logic for enterprise data
- ●Built backend payment and transaction API workflows for payment initiation, transaction verification, status updates, retry handling, and PostgreSQL-backed transaction persistence across pending, successful, failed, and retried payment states
- ●Improved request throughput by 30% and reduced average response latency by 28% by optimizing PostgreSQL queries, adding indexes, and validating endpoints through unit testing, integration testing, Git workflows, and CI/CD checks
Software Developer (Freelance)
- ●Built a kids coding game platform in React.js, TypeScript, and SQL/NoSQL, along with a job posting website supporting hiring workflows
Education
Master of Science in Information Technology
GPA: 3.79
Relevant Coursework:
Bachelor's in Computer Science and Engineering
GPA: 3.71
Skills & Technologies
AI / ML Engineering
Programming Languages
Frontend Frameworks
UI, CSS & Design Systems
Backend & APIs
Databases
Cloud & DevOps
Backend & Systems
Testing & Debugging
Core Concepts
Certifications & Recognition
Certified Generative AI Engineer Associate
Industry certification covering RAG pipeline design, LLM application development, model evaluation, and production deployment of generative AI systems.
View CredentialGenerative AI Engineer
Applied coursework in prompt engineering, embedding models, chunking strategies, and building retrieval-augmented applications end to end.
View CertificateMachine Learning for Natural Language Processing
Graduate-level AWS Academy track covering NLP model training, text processing pipelines, and managed machine learning services.
View BadgeAWS Cloud Architecting
Graduate-level AWS Academy track covering cloud architecture patterns, scalability, high availability, and secure infrastructure design.
View BadgeCampus Partner, 2025
Selected as an ASU Campus Partner for Perplexity AI, promoting AI-powered research tools and leading campus engagement for the Comet browser.
Featured Projects
- ●Built a platform that checks whether the skills on a resume are actually backed by code: it parses an uploaded PDF resume, deep-scans the candidate public GitHub repositories, and scores every claimed skill as proven or unproven, detecting 25+ skills across languages, frameworks, databases, and DevOps tools
- ●Designed a hybrid scoring engine that blends GitHub Linguist language statistics with code-pattern recognition to assign a proficiency level of Beginner, Intermediate, or Expert, processing repositories in parallel with early termination and prioritising config and entry-point files while excluding generated, minified, and bundled code; OpenAI GPT-4o mini then drafts targeted project plans for the weakest skills
- ●Front end built with Next.js 16 App Router, React 19, and TypeScript, using a CSS custom-property token layer to drive reusable card, button, and chip components, with animation cost controlled through shared keyframes, GPU-friendly transforms, and passive scroll listeners; back end runs on Next.js route handlers calling the GitHub REST API through Octokit, with PDF.js loaded from CDN so resume parsing never enters the bundle
- ●Built a research assistant for reading 10-K and 10-Q filings that answers questions such as comparing the AI-related risks of two companies, pulls the relevant passages straight from the filings, cites every claim with ticker, form, and filing date, and fact-checks its own answer before showing it
- ●Implemented the agent as a LangGraph state machine that plans sub-questions, retrieves, grades whether the evidence is sufficient, and loops back for the specific missing piece when it is not, capped at three retrieval rounds; multi-company questions retrieve each company separately so one cannot drown out the other, a verification pass strips any claim the source text does not support, and unseen companies are fetched from EDGAR and indexed on demand in around thirty seconds
- ●Front end is a dependency-free HTML, CSS, and JavaScript single page built on a CSS custom-property token layer with no framework or build step; back end is a FastAPI service deployed as a Vercel Function, running Llama 3.3 70B through Groq with a one-env-var swap to a local Ollama model, BAAI/bge-small-en-v1.5 embeddings computed locally, and Chroma as the persisted vector store
- ●Built a website for a Christian ministry where a visitor describes what they are feeling or wrestling with and receives a true Bible story, told warmly and applied to what they wrote, with every verse quoted from the King James text that ships with the app
- ●Designed the pipeline so the application does the Bible work before any language model is involved: a safety gate intercepts crisis, abuse, and out-of-scope messages and returns prepared responses carrying real help lines with no model call at all, recognised phrases match straight from a curated story library, and anything vaguer goes to semantic search across the whole KJV stored as vectors in Pinecone, leaving Groq responsible only for shaping retrieved material rather than sourcing scripture; answers are cached back into Pinecone so repeat matches skip the model entirely, and the curated library serves as a silent fallback whenever Groq is unavailable
- ●Front end built with Next.js 15, React 19, and TypeScript using hand-written CSS rather than a framework, with a custom-property token layer for theming, clamp-based fluid typography, and an animated story player; back end runs on Next.js route handlers integrating the Pinecone vector store and Groq, with a one-time ingest script that loads the King James text and curated stories into a serverless Pinecone index
BloomVe - People-First Job Sharing Platform
- ●Built a collaborative platform for sharing job opportunities inside private rooms, supporting two modes in Chats and Channels, and reaching MVP-plus status with full authentication, privacy controls, owner-managed permissions, and an invite system
- ●Engineered a companion Chrome Extension on Manifest V3 with a side panel, content scripts, and an auth bridge that shares the signed-in session with the web app, letting a user push a listing from any webpage into a room in one click, with Cheerio parsing page metadata server-side and XLSX export for taking a room list offline
- ●Front end built with Next.js 16, React 19, and TypeScript styled in Tailwind CSS 4, carrying a glass-morphism design language consistently across both the web app and the extension surfaces; back end runs entirely on Firebase using Auth, Firestore, and Storage, with firebase-admin on server routes and Firestore security rules enforcing room-based access control
Medical Inventory ERP - Full-Stack Inventory Management System
- ●Built an end-to-end ERP for managing medical supplies, covering inventory records, low-stock reorder workflows with priority classification, expiry tracking, category-level analytics, and printable reorder reports with an authorised signature block
- ●Implemented a reorder flow that walks a user through each selected low-stock item in turn with an automatically calculated suggested order quantity they can override, alongside a read-only reporting module serving backend-computed summary metrics, and pushed search, filtering, sorting, and pagination down into custom JPQL queries so the database does the work rather than the application
- ●Front end built with Angular 21 and Angular Material with route guards and an HTTP interceptor attaching the bearer token to every request, styled in hand-written CSS across four breakpoints with fluid typography, analytics charts drawn with runtime-computed CSS conic-gradient instead of a charting library, and dedicated print stylesheets driving the PDF reports alongside XLSX export; back end is a layered Spring Boot 3.5 service on Java 17 with JWT authentication, BCrypt password hashing, validated REST endpoints, projection DTOs for pre-aggregated reports, and JUnit and Mockito coverage including an end-to-end test of the JWT filter chain
AI-Powered API Performance Bottleneck Analyzer
- ●Built a service that reads production API logs and reports where the system is actually slow, flagging slow endpoints, high error rates, and database-heavy N+1 patterns, then explaining the likely root cause and the fix, cutting manual log analysis time by 80%
- ●Implemented P95 latency analysis with tuned detection thresholds at 500ms for slow endpoints and 5% for error rates, and made the AI provider swappable through a single environment variable so the same deployment can run against OpenAI GPT-4 or Anthropic Claude
- ●Back-end-only service with no custom front end: a FastAPI and Uvicorn application on Python 3.11, consumed through its generated interactive OpenAPI documentation, containerised with Docker and Docker Compose for portable deployment across ECS and Cloud Run
Budgetron - Mobile FinTech Application
- ●Built a cross-platform mobile application for personal expense tracking and splitting group payments, owning the work end to end from data model through API design to the shipped interface
- ●Reduced manual financial tracking effort by 40% through automated Excel and PDF report generation, and improved API response performance by 25% by optimising MySQL queries and tightening endpoint design
- ●Front end built with React Native in TypeScript; back end is a Node.js REST API over MySQL, with the schema and endpoint architecture designed around expense records and group settlement
Interactive Gamified Learning Platform
- ●Built a browser-based educational game that teaches phonics through interactive play, deployed as a live proof of concept that runs entirely in the browser with nothing to install
- ●Optimised the canvas render loop and game state management to hold a steady 60 FPS, and improved interface responsiveness through a tightened component architecture and careful event handling
- ●Front end built with React.js driving an HTML5 Canvas render loop; no back end, shipped as a fully static client-side build deployed on Vercel
Get In Touch
I'm always open to discussing new opportunities, collaborations, or just connecting with fellow developers. Feel free to reach out through any of the channels below!
Location
Open to relocation
FL / AZ