Open to Software Engineer roles · USA

Prajwal Narayanaswamy I build AI-powered, cloud-native, accessible, secure software - from the scraper to the screen.

Software engineer with 2+ years shipping production web apps, LLM features and data pipelines. Most recently at Microsoft World’s Edge, where I built pricing intelligence across 92 markets and review analytics over 100K+ player reviews.

prajwalnarayanaswamy8@gmail.com · linkedin.com/in/prajwal-narayanaswamy-2025041a0 · github.com/prajwaln18tech · prajwalnarayanaswamy.com

See selected work Resume (PDF)

Prajwal Narayanaswamy in a dark suit with arms crossed

~/prajwal — zsh

92markets priced
100K+reviews analyzed
29languages supported
216automated tests
25REST endpoints
2+years in industry

01 About

Engineer first. Product‑minded always.

Portrait of Prajwal Narayanaswamy — MS Computer Science Cal State Chico · 2025

I like the full journey of a feature: pulling messy data from the outside world, shaping it with solid APIs, and putting it in front of people through fast, accessible interfaces.

At Microsoft’s World’s Edge studio I turned two internal ideas into production platforms — one that tracks game pricing across Xbox, Steam and PlayStation in 92 markets, and one that uses embeddings and Azure OpenAI to make sense of 100,000+ player reviews in 29 languages. Both are secured behind Entra ID, covered by automated tests, and shipped through CI/CD.

Before that I built ETL pipelines at Cognizant and data tooling for research at Cal State Chico. I care about the unglamorous parts too: security regressions, WCAG compliance, and caching that makes cold starts disappear.

Focus: Full-stack · GenAI / LLMs · Cloud
Core stack: Python, FastAPI, React, TypeScript, Azure
Experience: 2+ years in industry
Motto: “Excellence in action is yoga.”

02 Experience

Where I’ve shipped.

Sep 2025 — Sep 2026

USA
Web Developer @ Microsoft — World’s Edge

via Experis
World’s Edge Price Tool Regional game-pricing intelligence
- Built an end-to-end platform (FastAPI + React 19/TypeScript) pulling prices from Xbox, Steam and PlayStation across 92 markets and 47 currencies, normalized to tax-free USD and benchmarked against US pricing with Excel reporting.
- Engineered resilient scrapers — Xbox Display Catalog API, Steam API, concurrent parsing of 68 PlayStation locales — with exponential back-off, a Selenium fallback, and an async job queue with rate limiting.
- Deployed on Azure App Service behind Front Door (WAF) with Entra ID sign-in; authored Bicep IaC for Container Apps and CI/CD in GitHub Actions and Azure DevOps.
- Fixed a reported production vulnerability with fail-closed Entra ID/API-key auth on every endpoint plus a 48-case regression suite; reached WCAG 2.1 AA with Playwright + axe-core across 6 app states and 5 screen sizes.
World’s Edge Review Tool AI-powered Steam review analytics
- Migrated a single-user Windows desktop tool into a multi-user web app with 25 REST endpoints, analyzing 100,000+ reviews in 29 languages — adopted by teams beyond the studio.
- Built multilingual semantic search on Sentence-Transformers with a content-addressed vector store; replaced ~227K per-pair calculations with one vectorized matrix operation.
- Designed automated insight detection and hybrid topic discovery (semantic matching + MiniBatchKMeans + Azure OpenAI labeling).
- Integrated Azure OpenAI for grounded RAG Q&A, bilingual reply drafting and theme analysis with prompt-injection safeguards; 216 pytest tests and single-flight caching that halved cold-start work.
- Python
- FastAPI
- React 19
- TypeScript
- Azure OpenAI
- Entra ID
- Front Door
- Bicep
- Docker
- Playwright
Feb 2024 — Oct 2024

Chico, CA
Instructional Student Assistant @ California State University, Chico
- Collected and analyzed research data for faculty, applying machine learning and statistical methods.
- Built reproducible preprocessing pipelines in Python (Pandas, NumPy) for cleaning, normalization and transformation.
- Turned raw research data into clear visualizations that communicated trends and results.
- Python
- Pandas
- NumPy
- scikit-learn
- Data viz
Nov 2021 — Jul 2022

Bengaluru, India
Programmer Analyst Trainee @ Cognizant Technology Solutions
- Developed complex ETL workflows in Informatica PowerCenter loading large datasets into data warehouses.
- Cut load times with session partitioning, pushdown optimization and parallel processing.
- Integrated flat files, Oracle, SQL Server and mainframe sources with SQL validation and cleansing rules; ran root-cause analysis on pipeline failures.
- Informatica
- Oracle
- SQL Server
- SQL tuning
- Data warehousing

03 Selected work

Systems, not just screens.

A closer look at how a few of these are put together.

Microsoft · Internal platform

World’s Edge Price Tool

Regional game-pricing intelligence. Scrapes three storefronts, converts every price into tax-free USD, and shows where a title is over- or under-priced relative to the US.

92markets
47currencies
68PS locales in parallel
48security regression cases

FastAPI
React 19
Selenium
Async queue
Azure Front Door
Bicep

Xbox Catalog API Steam API PlayStation ×68

Async job queuerate limit · back-off · Selenium fallback

Normalize47 currencies → tax-free USD Benchmark vs US Excel reports

illustrative · time-compressed

+0.00sidlePress run to watch rate limits, back-off and fallbacks play out.

Microsoft · Internal platform · GenAI

World’s Edge Review Tool

Turns a flood of multilingual Steam reviews into answers. Semantic search, automatic topic discovery, and grounded Q&A so teams can ask “why did sentiment drop in Brazil?” and get an answer grounded in the actual reviews.

100K+reviews
29languages
227K→1pairwise ops → one matrix op
½cold-start work

Sentence-Transformers
RAG
Azure OpenAI
MiniBatchKMeans
Recharts
pytest

hovertap a point · sample reviews are illustrative

Research · Accessibility · GenAI

Peer-reviewed JCSC 41(4) · 2025 · First author

Accessible Code Narrator

Screen readers such as JAWS read code word by word, top to bottom, and lose its structure. This VS Code extension narrates C++ so that the structure comes through: engineered prompts have an OpenAI LLM explain every line in context, such as which loop it belongs to, how many times it runs and what a condition decides.

A Python pipeline gives it two ElevenLabs voices. For each line, Rachel reads the code exactly as written, then Andy explains it, and ffmpeg stitches the pairs into one track.

2voices: code vs. context
95%code-to-speech accuracy
1stauthor, 5-person team
pp. 77–89Journal of Computing Sciences in Colleges

TypeScript
Python
VS Code API
OpenAI GPT-4
Prompt engineering
ElevenLabs TTS
ffmpeg

Read the paper ↗ Source code on GitLab ↗

binary_search.cpp Rachel · codeAndy · context

Real output

Hear what the tool actually produced: for every line, Rachel reads the code and then Andy explains it. Press play, or tap any line.

Real-time · Cloud-native

Multiplayer Chess Platform

Real-time chess over WebSockets with Django Channels, deployed on GCP Kubernetes and built to hold up under hundreds of concurrent games.

<100ms latency · 500+ concurrent users · 99.9% uptime

Django Channels
WebSockets
GKE
Kubernetes

04 Ask my résumé

Don’t skim. Ask.

A small retrieval engine over my résumé that runs in your browser — the same idea behind the Review Tool, scaled down. No server, no API keys, no tracking.

Pick a suggestion or type a question. Turn on semantic mode to download a 23 MB sentence-embedding model once and match on meaning, not just keywords.

Queryyour question
Tokenize · expandstemming + synonyms → BM25
EmbedMiniLM-L6 · 384-d · in-browser
Rank0.7·cosine + 0.3·BM25
Top-kpassages + sources

05 Stack

Tools I reach for.

01 Languages

Python
TypeScript
JavaScript
C++
SQL
HTML
CSS

02 Frameworks

FastAPI
React 19
Vite
Tailwind CSS
Recharts
Node.js
Django Channels
Pandas
NumPy
Selenium

03 AI / ML & LLMs

Azure OpenAI
GPT-4
RAG
Prompt engineering
Embeddings
Semantic search
Sentence-Transformers
scikit-learn
NLP

04 Cloud & DevOps

Azure App Service
Container Apps
Front Door
Entra ID
GCP
Docker
Kubernetes
GitHub Actions
Azure DevOps
Bicep
Linux

05 Data & ETL

Informatica PowerCenter
Oracle
SQL Server
MySQL
SQLite
Data warehousing
ETL tuning

06 Quality & Security

pytest
Playwright
axe-core
E2E testing
Regression testing
WCAG 2.1 AA
AuthN / AuthZ

06 Research, education & recognition

Foundations & milestones.

Publication · Journal article

AI-Assisted Contextual Code Narration for the Visually Impaired

Prajwal Narayanaswamy, Abbas Attarwala, Paul Viotti, Jaime Raigoza, Ed Lindoo

Journal of Computing Sciences in Colleges, 41(4), 77–89 · September 2025 · Consortium for Computing Sciences in Colleges · ACM Digital Library

Abstract

Conventional screen readers vocalize code linearly, word by word, and often fail to convey the contextual relationships and structural hierarchy that are fundamental to program comprehension. The paper proposes a system that uses an OpenAI large language model, guided by carefully engineered prompts, to turn source code into structured, context-aware descriptions. For example, it summarizes a deeply nested conditional before detailing each condition within that context. Implemented as a Visual Studio Code extension, it uses ElevenLabs text-to-speech with two distinct voices: one reads each line verbatim, and the other provides contextual explanations. The approach is designed to improve comprehension for blind and visually impaired programmers. It has not yet been formally evaluated with BVI participants.

View on ACM DL ↗ Source code ↗ Hear the technique in action →

Aug 2023 — Aug 2025

MS, Computer Science

California State University, Chico

Coursework: Algorithms, Data Science, Cybersecurity, Digital Forensics, Web Technology, Computer Networks.

Aug 2017 — Jun 2021

BE, Electronics & Communication

Visvesvaraya Technological University

Certifications

Google Cybersecurity ProfessionalCoursera
Python Data StructuresUniversity of Michigan · Coursera
IT Project ManagementCoursera

3rd Prize NASA Ames Space Settlement Design Contest
Semifinalist DST & Texas Instruments India Innovation Challenge, 2019

07 Contact

Let’s build something worth shipping.

I’m looking for my next software engineering role — full-stack, AI/LLM or cloud platform work. The fastest way to reach me is email.

Send an email LinkedIn GitHub

08 Under the hood

This page, measured.

Real numbers from your visit, read from the browser’s Performance APIs — not a screenshot of a Lighthouse score.

Page load: —
Time to first byte: —
Largest contentful paint: —
Cumulative layout shift: —
Page weight: —
Requests: —
First-party JS: —
Service worker: —

How it’s built

Zero frameworks. Semantic HTML, modern CSS and small native ES modules. No bundler, no build step.
Private by default. No analytics, no cookies, no third-party trackers.
Accessible. Skip link, visible focus, ARIA live regions, full keyboard control, and it respects reduced-motion.
Offline-ready. A network-first service worker keeps the page available without a connection.
On-device ML. Semantic search runs Transformers.js on WebAssembly, and passage vectors are cached by content hash.

View source ↗

Prajwal Narayanaswamy I build AI-powered, cloud-native, accessible, secureAI-poweredcloud-nativeaccessiblesecure software - from the scraper to the screen.