Skip to content

Software Engineer at Raaz

Hi, I'm Rudra.

Ask me anything.

This portfolio is a conversation. I'm an AI answering as Rudra. Ask about my work, projects, and background. Type a question, or tap one below.

Try asking

AI-generated — may be imperfect. Email rudrap5201@gmail.com for the real me.

[ Prefer to read? The full version ]

above: model-generated · below: human-written

Q.01aboutsource: human

Who am I?

I lead the stack behind a healthcare platform now past 55,000 patients: a React Native app, an Express backend, a Supabase database, and the boring glue that keeps all three in agreement.

Before Raaz I worked on browser-side ML at GJ-Map, compiling C++ inference kernels to WebAssembly so an ONNX model could run on a government analyst's laptop without a server round-trip. Before that I built a real-time fraud dashboard at SuperPe that flagged anomalous transactions before settlement.

Outside work I play Ultimate Frisbee. My team won gold at the National Open Championship in 2025 and the National College Championship in 2023, a useful reminder that latency matters off-screen too.

education
B.E. Computer Science
BITS Pilani, South Goa, Goa · Class of 2023
shipping for
3+ yrs
based in
Bangalore
India
reading
Designing Data-Intensive Applications
by Martin Kleppmann
Q.02experiencesource: human

Where have I worked?

Jan 2025 - Present · Bangalore, India

Software Engineer — Raaz

  • Primary engineer on the platform: the Express/TypeScript backend, the patient app, the assessment funnel, the browser-calling core of the rep console, and the satellite services around them (payments and fulfillment, appointments, notifications, analytics ETL), built and operated across two platform generations in eighteen months, alongside the engineers who own the doctor dashboards and call-center tooling. The patient base has grown past 55,000 registered, with 35,000+ monthly active.
  • Built the order pipeline that routes 120+ daily orders from app, web, and WhatsApp into one payment-status state machine that every channel's webhooks funnel into: Razorpay, Shopify, Shopflo, COD. Fulfillment runs through Prozo's warehouse and last-mile network; payment status only moves forward, so a late 'failed' can never overwrite a 'captured'. Idempotency is enforced at the database with a central webhook log and per-stream unique constraints, and cron reconcilers sweep for drift: a paid order missing its CRM deal gets found and repaired automatically.
  • A points-based clinical engine on validated instruments (IIEF-5, IELT bands) scores eight weighted root causes, stages ED/PE severity, and assigns a RED/AMBER/GREEN risk lane; a pure-function cart engine then drafts the first-month treatment plan, diffs doctor-corrected runs against the assessment draft, and pre-fills the doctor's record. It cut per-patient prep from 30 minutes to 10-15. All the math is in code; Claude and Groq only turn computed results into the doctor-facing report and patient copy, after an earlier version let the LLM do the arithmetic and it got the totals wrong.
  • Worked across three telephony generations in eighteen months: operated and extended the Knowlarity-era call-ops platform (Vapi and Smallest AI bot calls, outcomes auto-assigned to idle agents on a live console) and the Tata SmartFlo auto-dialer with its drain-aware campaign rescheduler, then built today's Plivo WebRTC browser softphone with per-agent SIP endpoints, a four-phase dial-state machine built from a measured ringback failure, and a call ledger reconciled against provider CDRs. It has handled 200,000+ calls a month at peak, with follow-up state derived, not accumulated, into Zoho CRM throughout.
  • Built server-side ad attribution (Meta CAPI, Google offline conversions) threading each click through to payment with pixel deduplication and DB-level idempotency, improving measured paid-ad attribution by 46%. Also built the bidirectional Zoho CRM sync, the appointment system, the notification system, and the e-commerce layer.
  • React Native
  • TypeScript
  • Express
  • Supabase
  • AWS
  • Zoho CRM
  • Claude API

May 2024 - Oct 2024 · Udaipur, India

Software Engineer — GJ-Map Solutions

  • Built six production React apps on the ArcGIS SDK with the Calcite design system for government and enterprise GIS teams working with heavy multi-temporal satellite imagery. The apps put years of imagery of the same land side by side so development becomes visible: roads appearing and extending, cropland shifting, construction filling in over time.
  • As one part of that, moved object detection into the browser: compiled C++ inference kernels to WebAssembly and paired ONNX Runtime with Meta's Segment Anything Model, cutting detection turnaround by ~40% versus the server round-trip and removing the server-side GPU dependency entirely.
  • Built the company's Node.js and SQL backend from scratch, including the geospatial ingestion, validation, and email-based submission-tracking pipelines that fed every map, plus the company website.
  • React
  • WebAssembly
  • ONNX Runtime
  • Node.js
  • ArcGIS SDK
  • SQL

Jul 2022 - Dec 2022 · Bangalore, India

Software Engineering Intern — SuperPe

  • Built a fraud-detection dashboard in React that flagged anomalous transactions in real time so support could intervene before settlement.
  • Built user onboarding and payment checkout in React Native, with geolocation-based service gating, in the payments checkout flow.
  • React
  • React Native
  • Node.js
  • PostgreSQL
Q.03projectssource: human

What do I build on the side?

Work systems live in the answer above. These are mine: open-source, with code you can read.

Open-Source Job-Search Agent

Scout

A job-search agent (MIT) I built during my own job hunt and open-sourced: plain TypeScript, no framework, a hand-rolled CLI and HTTP server. It pulls real openings straight from ATS APIs (Greenhouse, Lever, Ashby), HN 'Who is hiring', and RSS feeds, not job boards. It scores them with a zero-dependency heuristic plus an optional LLM pass, and drafts cover letters and cold emails with MX-validated address guessing. Thesis: zero ghost jobs.

  • TypeScript
  • Anthropic SDK
  • CLI
  • ATS APIs

Map-First Music Discovery

MoodMusica

Tap anywhere on a world map, type a mood, and get an AI-curated color palette and eight real, region-appropriate songs with playable previews (Lagos + 'energetic night out' → Afrobeats). Next.js with a hexagonal architecture (six swappable ports) and an anti-hallucination pipeline: every AI-suggested track is grounded in real Apple charts and verified against the iTunes API before it's shown. Free services only, no Spotify.

  • Next.js
  • React
  • Leaflet
  • LLM
  • iTunes API

Chat-to-Calendar Assistant

WhatsApp Calendar

A WhatsApp bot (Bun + Hono) that turns "meeting with Shivay tomorrow at 2pm" into a Google Calendar event with a Meet link and conflict detection. Groq parses intent into Zod-validated payloads; a hexagonal architecture with six port interfaces makes swapping providers a one-file change; and it does self-service Google OAuth over a sign-in link the bot sends you.

  • TypeScript
  • Bun
  • Hono
  • Groq
  • Google Calendar API
  • Fly.io
Q.04skillssource: human

What do I work with?

Languages

PythonTypeScriptJavaScriptC/C++SQL

Frontend

ReactReact Native (Expo)Next.jsTailwind CSS

Backend

Node.jsExpressPostgreSQLDynamoDBMongoDBRedisFirebase

Cloud & DevOps

AWS LambdaS3API GatewayEventBridgeCloudWatchDockerSupabaseVercelFly.io

AI / ML

LLMs (Claude, Groq, GPT-4o-mini)Voice AI (Vapi, Smallest AI)RAGONNX RuntimeSegment Anything ModelWebAssemblyYOLO

Integrations

RazorpayShopifyZoho CRMMeta CAPIGoogle Ads APIWhatsApp Cloud APIPlivoTata SmartFloPuppeteer
Q.05contactsource: human

How do we talk?

Email is the best way to reach me. I read everything and usually reply within a day. Hiring, collaboration, or a sharp question: all welcome.

Open to new roles