Jayesh Bhade — Portfolio
Profiling device · compiling shaders
0
FPS -- FRAME -- MS DRAW -- HEAP -- TIER -- DPR -- GPU --

I architect and ship real-time systems — from the socket to the pixel.

Fullstack engineer, weighted to the front. Trading-grade interfaces, sub-100ms budgets, platforms owned 0→1.

JAYESH BHADE
≤100msP90 webview load
5M+conversations served
0→1 ×4platforms shipped
Scroll
01 — SystemsFIG. 01 · N-body convergence

Architecture is deciding
where every part belongs.

A support platform is 30+ components, a socket layer, retry queues and a decision-tree engine. The discipline is making thousands of parts read as one system — move your cursor through the field; every particle knows its place and returns to it.

particles convergence 0.00 SCATTERED
02 — GeometryFIG. 02 · the math behind the motion

Animation is just
arithmetic on a clock.

01 / The claim

On the right, a square wave is being drawn by nothing but rotating circles. That is Fourier's theorem: any periodic signal — an audio buffer, a price series, an animation curve — is a sum of plain sines:

f(t) = Σ (4/kπ)·sin(k·t) // k = 1, 3, 5, … // one term = one circle: // radius 4/kπ // speed k rad/s

Each circle rides the tip of the one before it. The endpoint of the chain traces the signal — pure rotation in, geometry out.

02 / The transform

Your scroll position sets how many harmonics the chain carries. One circle is a lazy sine; two dozen snap into a square with the characteristic Gibbs ripple at the edges. Analysis is the same trick run backwards:

N(u) = 1 + ⌊u·23⌋ // u = scrollProgress Xₖ = Σₙ xₙ·e^(−i·2πkn/N) // the DFT

The DFT is the transform behind every audio visualizer — including the real-time pitch wave I shipped for Groww's voice tickets, drawn from the microphone as users speak.

03 / The easing

Linear motion reads as mechanical because nothing physical moves linearly. Every transition on this page routes time through an easing function first — a curve that reshapes the clock:

easeOutExpo(t) = 1 − 2−10t // fast start, asymptotic landing — // the same shape as a price settling // after a spike.

Performance engineering is this in reverse: knowing which milliseconds the eye will notice, and spending them deliberately.

N = 1 harmonics  kmax = 1
t = 0.00s  target = square(t)
fn = Σ 4/kπ · sin(k·t)
03 — Proofbenchmarks you can run

Speed is a feature.
Run it yourself.

// FIG. 03-A — webview cold start: stock vs my native-layer patch
Stock webviewbaseline
groww.in● interactive
0 ms
Preloaded webviewmy patch
groww.in● interactive
0 ms

A simulation of the production numbers: Groww's iOS webview went from a 1.8s P90 cold start to ~100ms after I moved preloading into react-native-webview's native layer. Same order of magnitude, felt here.

// FIG. 03-B — 915.trade: every surface is a live, draggable widget
915 WORKSPACE · SIMdrag the widgets
⠿ NIFTY 50 · 1s
⠿ STRADDLE · 24200
⠿ SCALPER · TAPE

The 915.trade thesis, reproduced: a pro terminal is not a page, it's a workspace — straddle premiums, scalper tape and index candles as independent live widgets, each draggable, each on its own feed. Tick rates adapt to your device tier.

// FIG. 03-C — Journey Builder: the graph that runs Groww's support chatbot
JOURNEY BUILDER · CANVASreact-flow style — drag nodes, edges follow
trigger
user messageintent: order_status
input
collect order idvalidation: regex
http request
GET /v1/orders/:idtimeout 800ms · retry ×2
switch
order.status3 branches
display
order timeline cardcomponent: OrderCard
http request
POST /v1/ticketsescalate → agent

A working miniature of the platform I built at Groww: ops compose support journeys from typed nodes — http request, display, switch, input — validated with graph algorithms (cycle detection, reachability) and compiled into the decision tree the chatbot walks. The production canvas, built on React Flow, powers 6 major workflows and 30+ contextual answers across 5M+ conversations.

04 — Selected workproduction standards, everywhere

Systems I've
designed and owned.

Shipped · 915.Trade — by Groww

915.trade

F&O widget terminal

An advanced futures & options terminal built from zero for professional traders — straddle charts, scalper tape and index views as live, draggable widgets, where rendering latency is the product. See FIG. 03-B above for a working miniature.

Shipped · Groww

Support Platform

Chatbot + Journey Builder

Architected the chatbot serving 5M+ conversations — 30+ components, WebSockets, AI streaming, retry semantics — and the graph-based Journey Builder that lets ops author its decision trees. Packaged as an SDK, now embedded in three apps.

Shipped · Keychain

Keychain OS

ERP for manufacturers

Building the frontend of an ERP-style operating system for manufacturers — monorepo architecture, design system, and the data-heavy surfaces that make an ERP feel fast.

05 — Capabilitiesspec sheet

Fullstack, weighted
to the front.

Languages
TypeScriptJavaScriptHTMLCSSSQLGLSL
Frontend
ReactNext.jsReact Native · Android webviewReact FlowCanvas & WebGL
Backend
Node.jsNest.jsExpressWebSocketsPostgres · pgvectorpg-boss queues
State & data flow
ReduxZustandreal-time streamingAI/LLM integration — streaming, routing, cost guards
Architecture
Frontend system designmicro-frontendsmonorepos — Turborepo · pnpmwebview SDKsgraph algorithms
Practice
Performance engineering — sub-100ms budgets0→1 ownershipnative-layer patches when the web runs outAI-accelerated development
06 — AI Workflowsthe agentic toolchain

One engineer,
run like an org.

Every ADR, issue and PR in TheBrain moves through Claude Code skills I wrote for the repo — a PM lens, an architect lens, a security lens, and an autonomous fix-issue loop, with me as the reviewer of record. Alongside them, a set of meta-workflows adapted from David Ondrej's open skill library keeps the agents themselves honest.

IDEA/grill-me/pm · /architect · /security-issue/batch-github-issues/preflight/fix-issue/ship/adrMERGED
RUNBOOK 06-Aissue pipeline — written for TheBrain
/pm

Files issues as a product manager — user story, acceptance criteria tied to user value, success metrics.

/architect

Files design issues — architectural impact, data-model changes, service boundaries, perf and scale questions that must be answered before code.

/security-issue

Security review lens — threat model, attack surface, mitigations written as acceptance criteria.

/spike-issue

Time-boxed investigations where the deliverable is a decision and a document, not shippable code.

/batch-github-issues

Turns a plan into a sequenced backlog — priority/size/scope labels, milestones, blocked-by edges wired between tickets.

/preflight

Reconciles the true state of repo, tracker and permissions before any work — main sync, in-flight audit, stale-lock detection. Multiple agents work the repo; never trust a stale snapshot.

/fix-issue

Autonomously takes a ticket to a merged PR — claim, implement test-first, verify, open PR, merge, update labels.

/ship

Lands a PR on main around the repo's real merge quirks — disabled auto-merge, non-required checks, stacked PRs that fold into their base.

/test-doctor

Diagnoses environmental test failures — shared test-DB drift, bare worktrees, wall-clock flakes — before anyone debugs phantom code bugs.

/adr

Records architecture decisions permanently — what, why, alternatives, consequences — so nothing gets relitigated.

/dev

Boots the full local stack in order: Docker services, health checks, migrations, then every app via Turbo.

RUNBOOK 06-Bmeta-workflows — adapted from davidondrej/skills
/grill-me

The agent interviews me relentlessly about a design, one question at a time, until no branch of the decision tree is unresolved.

/brain-to-docs

A Q&A loop that extracts vision, taste and decisions out of my head into README and ADRs.

/handoff

Compacts a working session into a single brief a fresh agent can continue from — no re-discovery, no repeated mistakes.

/delegating-to-agents

Agent-to-agent orchestration — picking the right agent for a subtask, prompting it, polling its progress.

/agent-self-scheduling

Puts agents on a clock — cron, heartbeats, recurring autonomous checks.

/deep-research

Rigorous one-paragraph research prompts fired at a research API, ending in a cited markdown report.

/effective-agent-skills

The meta-skill — how to author skills that trigger reliably: anatomy, progressive disclosure, anti-patterns, testing.

07 — Journey2019 — present

Indore → Bangalore
→ Gurugram.

2019 — 2023IET DAVV, Indore
B.E. Engineering

Foundations

Technical Head of the Entrepreneurship Cell. Shipped the institute's ACIIE website in my second year — the first time strangers depended on my code. Built a MERN commerce platform, a React Native Covid-19 tracker, and a GPT-powered voice assistant before that was a product category.

ReactReact NativeMERNOpenAI API
Jul 2022 — Jun 2023Groww, Bangalore
SDE Intern

First production surface

Shipped voice messages for support tickets with real-time pitch-wave visualization rendered as users speak. Over 20% of Groww's support tickets are now voice-based. Rebuilt Help & Support for Lighthouse and SEO; designed the Recent Activities module.

Web Audio APICanvasSEO
Jun 2023 — Apr 2025Groww, Bangalore
SDE 1

Owning a platform end to end

Architected the customer-support chatbot from scratch — 30+ reusable components, WebSockets, AI streaming, message queuing and retry semantics. 5M+ conversations served, with a measurable lift in self-serve resolution.

Designed Journey Builder on React Flow — graph algorithms powering 30+ contextual answers and 6 major workflows across Stocks, F&O, Mutual Funds, IPO, UPI and Balance. The tool ops uses to program the bot without engineers.

When the web platform ran out, I went below it: patched react-native-webview's native code to preload webviews — iOS P90 cold start 1.8s → 100ms — then packaged the chatbot as an SDK now embedded in the Wealth and GrowwCredit apps.

WebSocketsReact FlowGraph algorithmsNative webviewSDK design
2024Groww, Bangalore
Internal Hackathon

Hackathon winner

Won Groww's 2024 internal hackathon with a real-time fundamental-analysis chatbot for stocks — live market data fused with AI reasoning, built in a weekend, demoed to the company.

Real-time dataAI
Apr — Aug 2025915.Trade — by Groww
SDE 2

915.trade — the pro terminal

Moved to Groww's pro-trading arm as SDE-2 to build its advanced F&O charting platform from zero — a widget-based terminal where straddle charts, scalper tape and index views are live, draggable surfaces, and where speed and real-time correctness are the entire value proposition.

F&OChartingWidget workspaceReal-timePerformance
Aug 2025 — PresentKeychain, Gurugram
Frontend Developer

Keychain OS

Building the frontend of an ERP-style operating system for manufacturers — architecture, design system, data-heavy surfaces. The largest canvas yet.

Next.jsTypeScriptTurborepoSystem design