Raj Joshi
RAJ JOSHI · SOFTWARE ENGINEER · MUMBAI<h1>the one line every section below must prove</h1>

I make complex systems visible.

Full-stack engineer at J.P. Morgan, Mumbai. Five years turning research and portfolio data into tools people can understand. Outside work, I build exchanges, cubes and AI agents you can see inside.

See the workNo time? Get my CV (PDF)LinkedIn · GitHub · Credly

Things I've built

flip "See inside" to open any project uparchitecture · decisions · bugs · numbers
Prayog screenshot
FLAGSHIP · MVP DONE

Prayog

A simulated stock exchange and bot arena. Simulated traders move the market; you trade by hand or with your own bot.

Java 21 · LMAX Disruptor · Kafka · PostgreSQL · Redis · React · Python SDK
How it worksCode
Prayog · inside
Traders + bots→Gateway→Disruptor ring→Matcher→Journal→Kafka→Ledger
why it exists
my 2021 induction win was an algo strategy for an exchange I could not see inside, so I built one I could
key decision
one single-threaded matcher, journal first: replay gives a byte-identical event log, and Kafka downtime never stops trading
hardest bug
p99 hit 445 ms with 3.5 s spikes. Traced it to Serial GC pauses in a memory-starved container; ZGC made it 5x worse, so I kept the default and wrote down why
numbers
~100 ns per command matching · 20,000 commands/s with fsync · live p99 14 ms · 23 ADRs
ASK ME ABOUT
Why an engineer at a bank builds his own exchange, and why the disk, not the code, sets the latency.
How it worksCode
Galois screenshot
LIVE · v0.6

Galois

Group theory through an interactive 3D Rubik's cube, from identity and permutations to why every position solves in 20 moves.

TypeScript · React · cubing.js · PWA · Vitest · Playwright
LiveCode
Galois · inside
Move→Permutation→Group op→Lesson check→3D render
why it exists
I learned the cube from YouTube at 13; this is the version that explains the math
key decision
turn first, name it later: every idea starts as something you do on the cube; fully static on GitHub Pages
hardest bug
a release that never arrived: the offline service worker kept serving the old app, so new workers now take over as soon as they install
numbers
6 of 8 phases shipped · tested on phone, tablet and laptop · Solve of the Day
ASK ME ABOUT
How the same group axioms explain Git, undo stacks and CRDTs.
LiveCode
Finfluencer AI screenshot
TOP 700 / 57,000+ · AGENTIC AI DAY

Finfluencer AI

A personal financial advisor that reads your real money data through Fi Money's MCP server and turns a goal ("open a dance studio", "retire by 50") into a plan. Built for Google Cloud Agentic AI Day, July 2025.

Gemini · Google ADK · Vertex AI · MCP · Streamlit
Demo video
Finfluencer AI · inside
Question→Orchestrator→Fi MCP tools→MoneyMind · Lakshya AI→Gemini→Plan
why it exists
people have the data about their money but no one to explain it
key decision
one orchestrator routes to specialist agents (MoneyMind for cash flow, Lakshya AI for goal plans) and asks for missing inputs before it plans
numbers
top 700 of 57,000+ developers · net worth, credit score, EPF, funds and stocks from one MCP server
ASK ME ABOUT
Designing agents that touch real money data: guardrails and context engineering.
Demo video

Journey

the stories behind each line
  1. May 2025 — now
    Associate, Software Engineer II
    J.P. Morgan · Portfolio Insights, Asset & Wealth Management

    I help deliver features for J.P. Morgan's Model Portfolios platform, which financial advisors use to choose, compare and present model portfolios to their clients. I take each feature from discovery with the product owner through to production.

    • Opened the platform to alternatives and separately managed accounts, a new business line expected to scale fast.
    • Built white-labeling, so an advisor's firm name and logo appear on every fact sheet and trade-commentary PDF.
    • Connected models to the Heatmap fund-comparison app, and designed a fund-suitability check that compares a fund against the client's existing portfolio.
    • Moved the platform to new EKS and EMR clusters with zero downtime, using blue/green deployments.
    2025–26 City lead & SME, AWS AI LeagueJul 2025 Top 700 / 57,000+, Agentic AI Day
    Stack: Java, Spring Boot, React, EKS/ECS, Kafka, DynamoDB, Redis. I use GitHub Copilot and Roo Code daily, with reusable agent skills, and mentor junior engineers.
  2. Jul 2021 — Apr 2025
    Analyst → Associate, Software Engineer
    J.P. Morgan · Index Research

    I owned a client-reporting application and the daily-operations modules behind it, working with the operations team to turn their manual tasks into software.

    • Redesigned the database schema that limited growth, then added caching, so new clients onboard easily and monthly reports reach a large client base on time.
    • Led the Java 8 → 11 → 17 upgrade and the iBatis → MyBatis migration, and removed an outdated Sybase package for faster, safer releases.
    • Built end-user feedback with Lucene search, a Cypress/Cucumber BDD proof of concept, and parts of a new ETF platform (FastAPI, PostgreSQL, AWS) that replaced a legacy system.
    2024 Co-led DeepRacer for 700+2023 2nd place, DeepRacer, led team of 62022 Won SEPathon hackathon2021 Won induction with an algo-trading app
    DeepRacer 2023: nobody on the team could tell why the car made certain decisions, so I built a tool that drew what our reward function was actually rewarding on the track. That is how we climbed to 2nd. The next year I co-ran the competition for 700+ people.
  3. 2017 — 2021
    B.E. Computer Engineering
    VESIT, Chembur · University of Mumbai
    Paper · Machine Learning and Technology in Diagnostic Research: The Case Study of Cognitive Disorders among Toddlers (SSRN, 2020)
  4. 2015 — 2017
    School · HSC and SSC
    Swami Vivekanand Junior College, Dahisar · St. Francis High School, Bhayandar
    Taught myself the Rubik's cube at 13–15 from YouTube. Ten years later it became Galois.

Skills

grouped by where I use themevery skill points to where it was proven
AI and agents
LLM agentsBedrockVertex AIGeminiGoogle ADKMCPMulti-agent designContext engineering
proven in · Finfluencer AI (top 700 / 57k) · AWS AI League city lead
Backend
JavaSpring BootPythonFastAPIDjangoKafkaRESTLucene
proven in · 5 years of production Java at J.P. Morgan · Prayog matching engine and Kafka ledger
Frontend
ReactReduxTypeScriptViteJavaScript
proven in · Galois 3D app · Prayog trading workspace · portfolio insights UIs at work
Data and cloud
AWS EKS / ECS / EMR / LambdaDockerTerraformPostgreSQLDynamoDBRedis
proven in · Prayog: 11-container stack, Prometheus and Grafana
Engineering practice
CI/CDJenkinsGitSonarCypressCucumber BDD
proven in · Galois: Vitest + Playwright in CI · Prayog: 23 decision records, benchmarks with dates
Finance domain
Asset managementModel portfoliosETFsPortfolio analyticsClient reporting
proven in · Index Research → Portfolio Insights · Prayog

Beyond work

what I do when nobody is paying mehow each interest shows up at work
COMMUNITY
Running AI competitions

City lead and SME for the AWS AI League in Mumbai; co-led DeepRacer for 700+ engineers.

shows up at work as · leading without authority, explaining ML to non-ML engineers
PUZZLES
Self-taught cuber

Learned the Rubik's cube from YouTube as a teenager; now I teach the math behind it.

shows up at work as · breaking hard problems into moves you can reason about
MARKETS
Markets and game theory

Strategy games, auctions and how real traders differ from rational ones.

shows up at work as · the instinct behind Prayog and its bots
MUSIC AND MIND
Learning piano and DJing

Deep music listener, learning to play and mix; reading psychology and how people learn.

shows up at work as · practice loops and spaced repetition in how Galois teaches