Open to Fall 2026 Co-ops

Arun Teja ReddyKallam

CS student at ASU. I build

Core Stack

PythonTypeScriptReactNext.jsNode.jsFastAPIClaude APILangGraphMongoDBSQL

Email me Resume

Press⌘Kto navigate anywhere

ASU ’27

B.S. Computer Science

Tempe, AZ

Greater Phoenix area

0Internships

0Projects shipped

0K+Rows analyzed

0+Profiles validated

Education

Arizona State University · Tempe, AZ

B.S. in Computer Science

Aug 2023 - May 2027

Relevant Coursework

Intro to Artificial IntelligenceData Structures & AlgorithmsSoftware EngineeringFoundations of Machine LearningDatabase ManagementMobile App DevelopmentOperating SystemsObject-Oriented ProgrammingPrinciples of Programming LanguagesInformation AssuranceHuman-Computer InteractionFoundations of Data VisualizationApplied Linear AlgebraProbability & Statistics

Academic Standing

Dean’s List

Graduating May 2027

4.0GPA

Focus Areas

AI / LLM Engineering

LangGraph agents, RAG, QLoRA fine-tuning, eval harnesses

Full-Stack Development

Web apps end-to-end, APIs, and databases

Data & Analytics

Pipelines, dashboards, and insight delivery

Skills

Tap or hover a category to explore depth across six domains

AI / LLM

LangGraphMCPRAGNLPQLoRAClaude APIOpenAI APIPyTorchHugging Face

Web & Frameworks

ReactNext.jsNode.jsExpress.jsFastAPIRedux ToolkitStreamlit

Languages

PythonJavaScriptTypeScriptSQLGraphQLJavaHTMLCSS

Data & Analytics

PandasNumPyscikit-learnTableauPower BI

Databases

MongoDB AtlasMySQLChromaDB (Vector)

Tools & Cloud

AWS LambdaAPI GatewayECRDockerGitVercelpytestShopify

Experience

Web Development Intern

BYLD CommerceLehi, UT

Jun 2026 - Present

JavaScriptSquarespaceShopifyPythonWeb Design

Redesigned and shipped the agency's marketing website end to end: a mobile-first, story-driven single-page site in vanilla JavaScript, injected into Squarespace to achieve a fully custom design on a no-code platform.
Built a Python build pipeline that cut site updates from full manual rebuilds to minutes.
Shipped front-end fixes, theme customizations, and UX improvements on client Shopify storefronts.

Junior Data Analyst

Food Forest AIPhiladelphia, PA

Jun 2025 - Jul 2025

PythonPandasData QualityQA Automation

Cleaned and validated 500+ food & beverage company profiles, turning raw scraped data into structured enrichment datasets feeding the production B2B search engine.
Automated data quality checks in Python (Pandas) to flag missing values, anomalies, and formatting errors, cutting manual review load and keeping the ingestion pipeline supplied with production-ready data.

Full-Stack Web Development Intern

Prodigy InfoTechMumbai, India

Sep 2024 - Oct 2024

MERNJWT AuthbcryptMongoDBReactExpress

Shipped a MERN authentication portal with JWT sessions, bcrypt password hashing, rate-limited endpoints, and role-based access control on React routes and Express middleware.
Built a full-stack Employee Management System with admin-only CRUD, server-side search/filter/sort, and a MongoDB aggregation pipeline dashboard surfacing real-time headcount, average salary, and department breakdowns.

Projects

6 case studies, every one shipped · 4 live demos · 1 published model. Scroll: on desktop the deck stacks as you go.

AI Support Agent for E-commerce

AI / LLMFull-Stack

AI customer support agent for a Shopify store: a LangGraph state machine over the Anthropic API with RAG retrieval, tool calling, and layered safety, live on AWS Lambda + Vercel.

PythonLangGraphMCPClaude APIRAGChromaDBGraphQLAWS LambdaVercel

53-case eval harness (deterministic graders plus a tool-grounded LLM judge), now the project's regression suite passing at 100%; its failure analysis drove architecture fixes and a measured model choice, 2.4x cheaper with no accuracy loss
Layered safety: a pre-filter blocks prompt-injection attacks before any model call, tools are allowlisted, and a grounding check rejects any order claim not backed by Shopify data
Custom MCP server exposing read-only Shopify Admin GraphQL tools with email-match authorization; plugs unchanged into any MCP host, including Claude Desktop

Live app Code

100%

53-case eval pass

$0.18

per 100 conversations

3.8s

p95 latency

CEFR QLoRA Benchmark

AI / LLMData

Does a QLoRA fine-tuned Llama 3 8B beat the GPT-4o-mini call my Feedback API makes? Yes, and the benchmark proves it with reproducible, honest evals.

PythonPyTorchQLoRA / PEFTTRLLlama 3 8BHugging Face

4-bit QLoRA fine-tune (transformers, peft, trl) designed to answer with a single token, eliminating output-parsing failures and cutting inference to 0.19s
Deterministic pipeline over 10k expert-labeled sentences: checksum manifests, leakage audits, and a single frozen test pass for reproducible, contamination-free evaluation
Honest ablations documented as negative results; fine-tuned adapter published on Hugging Face with a full reproduction recipe

Model on HF Code

Llama 3 8B QLoRA62.7%

GPT-4o-mini few-shot40.8%

Exact accuracy, same 1,460-sentence held-out test set

24x

lower cost per request

lower latency

99%

within one CEFR level

Goalsetter+

AI / LLMFull-Stack

Production-deployed MERN goal tracker with AI-powered SMART goal suggestions, a hand-built SVG analytics dashboard, sub-tasks, share links, and natural-language due dates.

ReactNode.jsExpress.jsMongoDB AtlasRedux ToolkitClaude APIJWT

Claude API generates SMART goal suggestions from plain-English input, rate-limited server-side, with one-click add
JWT auth, sub-tasks, natural-language due dates, overdue detection, and revocable public share links
Analytics dashboard hand-built in SVG from Redux state (13-week heatmap, streak tracking), plus a one-click demo account; deployed on Vercel and Render

Live app Code

SMART goal generator

13-wk

SVG heatmap + streaks

Live

one-click demo account

LLM Multilingual Feedback API

AI / LLMFull-Stack

FastAPI service with a built-in web app giving language learners structured feedback: minimal corrections, categorized errors explained in their native language, and a CEFR difficulty estimate.

PythonFastAPIOpenAI APIClaude APIPydanticDockerGitHub Actions

Pluggable LLM backends selected by environment variable: OpenAI, Anthropic, or a zero-config offline mode so the full system runs before any key is configured
Raw model output is never trusted: every response passes a normalization and validation layer before it reaches the client
Ships its own no-build web app: word-level correction diffs, categorized error cards with native-language explanations, shareable re-run links, and one-click examples

Code

pluggable LLM providers

languages in the demo

tests green in CI

NBA Teams Dashboard

Full-Stack

Editorial-style dashboard for all 30 NBA franchises: team pages, live search, and franchise stats with motion-rich UI.

Next.js 16React 19TypeScriptTailwind CSSFramer Motion

Next.js 16 App Router with async Server Components; self-hosted data layer (JSON snapshot + 30 PNG badges) after the public API became unreliable
Non-blocking live search with React 19 useDeferredValue, per-team accent colors, skeleton loading states, and route-level error boundaries
Flash-free dark/light theme toggle via a pre-paint inline script, plus an a11y pass: semantic landmarks, focus rings, ARIA live regions

Live app Code

franchises, full team pages

React 19

non-blocking live search

A11y

landmarks, focus, ARIA

Sales Insights Dashboard

Data

Interactive analytics dashboard over 94,073 AtliQ Hardware transactions, live on Streamlit Cloud and backed by a Railway-hosted MySQL database.

PythonMySQLSQLAlchemyStreamlitPlotlyPandas

MySQL pipeline with a sales_cleaned view normalizing multi-currency data, joining 5 dimension tables, and resolving 7 data-quality issues
10 KPI queries using window functions (LAG, SUM OVER PARTITION BY); surfaced one customer driving roughly 33% of INR 51.77 Cr revenue
Year/market/zone filters with a CSV fallback layer for production resilience; deployed on Streamlit Cloud + Railway

Live app Code

94K

transactions modeled

KPI queries, window functions

33%

revenue risk surfaced