SKIP TO CONTENT
third-year CS · Vellore Inst. of Technology · Software Engineering

Shourya
Upadhyaya.

Software Engineer & Distributed Systems
Software engineer specializing in distributed backend architectures, real-time audio & ML pipelines, and high-performance full-stack systems.

Résumé

Background

Profile

Third-year Computer Science undergraduate at VIT Vellore, based in Bengaluru, focused on backend-heavy web products, real-time streaming pipelines, and database-centric systems. I build production services with Node.js, TypeScript, Python, Go, PostgreSQL, and Redis, deploy them in Docker behind Nginx, and instrument them with Prometheus and Grafana.

Location

Bengaluru, India

Education

Vellore Inst. of Technology
B.Tech in Computer Science & Engineering
2024 — 2028

Selected Projects

7 / 7 PROJECTS
Production Full-Stack & Real-Time Architecture01 / 07 · Jan 2025 — Present

OpenPumTa

The 21-Day Deep Work & Study OS (300+ Users)

Architected 30+ REST APIs and 10 Prisma ORM models backed by PostgreSQL. Built a stateful WebSocket hub for sub-second cross-device Pomodoro timer synchronization, added Redis caching that reduced analytics database load by 80%, integrated AI daily habit reviews, and deployed a multi-container Docker Compose stack on a Linux VPS behind Nginx with Prometheus and Grafana observability.

300+ Users
10,000+ total platform visits
-80% DB Load
via Redis write-through cache
30+ APIs
10 Prisma models & WS sync
  • Next.js
  • React
  • TypeScript
  • Tailwind CSS
  • Zustand
  • Node.js
  • Express.js
  • REST APIs
  • WebSockets
  • PostgreSQL
  • SQL
  • Prisma
  • Redis
  • Docker
  • Nginx
  • Prometheus
  • Grafana
  • Linux
DOCKER COMPOSE · LINUX VPSWEB CLIENTNext.js AppSECOND DEVICECross-tab syncNGINX PROXYTLS · routingWS TIMER HUBstateful roomsEXPRESS API30+ REST · ZodREDIS CACHE-80% DB loadPOSTGRESQL10 Prisma models
Trace · Analytics & Habits

Habit completions and study logs pass through Nginx and Express (30+ REST APIs), update the Redis analytics cache (-80% DB load), and persist across 10 Prisma models in PostgreSQL.

Real-Time Audio ML & Streaming Systems02 / 07 · Sep 2025 — Dec 2025

VoiceShield AI

Real-Time Audio Scam & Deepfake Detection (SIH Finalist)

Engineered a Node.js and TypeScript WebSocket media gateway ingesting live 48 kHz mono PCM16 call streams, reassembling arbitrary network frames into ordered 3-second sliding windows while enforcing a strict byte-conservation invariant. Routed chunks over a persistent WebSocket to a FastAPI ML engine running RawNet2, WavLM, and SpeechBrain acoustic deepfake models alongside OpenAI Whisper STT, achieving ~1.1s median CPU latency while queuing PostgreSQL writes off the hot audio path.

~1.1s CPU
median inference per 3s window
48 kHz PCM16
100% byte-conserving ring buffer
SIH Finalist
off-path async PostgreSQL queue
  • Node.js
  • TypeScript
  • Python
  • FastAPI
  • PyTorch
  • OpenAI Whisper
  • ONNX Runtime
  • SpeechBrain
  • WebSockets
  • PostgreSQL
  • Prisma
  • Next.js
  • Tailwind CSS
STREAMING GATEWAY · FASTAPI ML48kHz PCM16CallVault · SIP3s RING BUFFERbyte-invariantACOUSTIC MLRawNet2 · ONNXWHISPER STTlive transcriptASYNC DB QUEUEoff-path writeSCAM AGENT1.1s median CPUSOCKET.IO UINext.js alert
Trace · Dual-Model Scam Trace

48 kHz PCM16 audio is assembled into 3-second byte-verified chunks, evaluated in parallel by RawNet2/ONNX and Whisper in ~1.1s on CPU, and streamed to the Next.js dashboard.

Backend Architecture & Payment Systems03 / 07 · Apr 2025 — Oct 2025

Moss X

High-Performance E-Commerce Modular Monolith

Built the full-stack e-commerce platform for 100+ registered users using React, Express.js, and MongoDB aggregation pipelines for sales analytics and product recommendations. Cached the catalog API in Redis to slash latency by 90% (1.5s to 150ms) and keep homepage load below 50ms. Integrated Razorpay with HMAC webhook signature verification and idempotency keys, processing 20+ live orders with zero duplicate charges.

1.5s → 150ms
catalog API latency (-90%)
<50ms Load
serving 100+ registered users
0 Duplicates
HMAC idempotent Razorpay webhooks
  • JavaScript
  • Node.js
  • Express.js
  • REST APIs
  • MongoDB
  • Redis
  • React
  • Zustand
  • Tailwind CSS
  • Docker
EXPRESS MODULAR MONOLITHSTOREFRONT<50ms React SPARAZORPAYsigned webhookAPI GATEWAYExpress · AuthREDIS CACHE1.5s → 150msWEBHOOK GUARDHMAC · idempotentMONGODBorders · catalogAGGREGATIONSsales analytics
Trace · Idempotent Payment

Razorpay webhooks are verified via HMAC-SHA256 and idempotency keys in the Express gateway, committed once to MongoDB, and rolled up into admin sales analytics.

Database Internals & Cloud Systems04 / 07 · Jan 2026 — Mar 2026

CRMS Cloud Control Plane

Database-Centric Cloud Resource & Docker Telemetry Engine

Designed a 14-table 3NF/BCNF PostgreSQL 16 control plane where multi-tenant isolation is enforced at the database engine via Row-Level Security (app.current_org_id) and btree_gist temporal exclusion constraints. Eliminated check-then-act VM quota race conditions using SELECT FOR UPDATE row locks inside transactional stored procedures, and partitioned a 1,000,000+ row telemetry table with BRIN & GIN indexes to cut time-range query latency from 42ms to 0.2ms.

42ms → 0.2ms
on 1M+ BRIN-partitioned rows
PG16 RLS
& SELECT FOR UPDATE quota locks
Live SSE
pg_notify Docker telemetry stream
  • PostgreSQL
  • SQL
  • PL/pgSQL & RLS
  • TypeScript
  • Node.js
  • Express.js
  • REST APIs
  • React
  • Tailwind CSS
  • Docker
  • Nginx
POSTGRESQL 16 ENGINE · 3NF/BCNFCLOUD CONSOLETenant API reqDOCKER PROXY30s containerRLS CONTEXTSET LOCAL org_idQUOTA LOCKSELECT FOR UPDATEINGEST PROCsp_ingest_metrics1M+ PARTITIONSBRIN · GIN 0.2msPG_NOTIFY SSElive alert push
Trace · 1M-Row Telemetry

Live Docker stats ingest via PL/pgSQL into 1M+ BRIN/GIN-indexed partitions (queried in 0.2ms vs 42ms) and push real-time alerts via pg_notify SSE.

Automation & Applied AI Pipeline05 / 07 · Feb 2026 — Present

Job Agent

Autonomous Job-Discovery, Multi-Resume Scorer & ATS Classifier

Built an Express 5 + Next.js 16 platform that ingests roles from Greenhouse, Lever, Ashby, and Workday, scores each posting against multiple LaTeX/PDF resumes using a deterministic 0–100 weighted rubric (+35 skills, +20 role, +15 domain) before optional LLM refinement, classifies recruiter emails via a 12-class Gmail OAuth state machine with forward-only pipeline transitions, and executes dry-run-first Playwright application workflows.

0–100 Rubric
deterministic multi-resume scorer
12-Class NLP
forward-only Gmail state machine
8+ ATS Feeds
with dry-run Playwright runner
  • Next.js
  • React
  • TypeScript
  • Tailwind CSS
  • Node.js
  • Express.js
  • REST APIs
  • PostgreSQL
  • Prisma
  • Playwright
EXPRESS 5 · PRISMA · PLAYWRIGHT8+ ATS FEEDSGreenhouse · LeverGMAIL OAUTHrecruiter threadsRESUME SCORER0–100 rubric12-CLASS NLPmonotonic statePOSTGRESQLPrisma stateNEXT.JS 16 UI7-tab workspacePLAYWRIGHTdry-run · SSE
Trace · Score & Auto-Apply

Postings from 8+ ATS boards are scored 0–100 deterministically against multiple resumes, stored in PostgreSQL, reviewed in Next.js 16, and executed via dry-run Playwright.

Concurrent Systems & CLI Tooling06 / 07 · Jan 2026 — Feb 2026

GoLakeScanner

Concurrent Data-Lake File Classifier & Indexer in Go

Engineered a dependency-inverted Go CLI and library that walks nested data lakes, computes streaming SHA-256 hashes to classify files as new, modified, or unchanged, and dispatches pending files through a generic bounded worker pool (pool.Run[In, Out]) with 30-second per-file context.WithTimeout isolation and signal-aware graceful shutdown (SIGINT / SIGTERM).

pool.Run[T]
generic bounded goroutine pool
SHA-256 Diff
64 KB streaming incremental skip
30s Timeout
per-file isolation & pgx/v5 catalog
  • Go
  • PostgreSQL
  • SQL
  • Linux
GO CONCURRENCY · POOL.RUN[IN, OUT]DATA LAKEPDFs · scansSHA-256 WALKER64KB stream hashFAST SKIPunchanged digestWORKER POOLpool.Run[In,Out]FILE PROBES30s ctx timeoutPGX/V5 CATALOGPostgreSQL state
Trace · Phase 2: Worker Pool

New and modified files flow from the SHA-256 walker into the generic pool.Run[In, Out] goroutine pool, run 30s timeout-isolated PDF/magic-byte probes, and commit to PostgreSQL.

High-Concurrency Frontend & State Engineering07 / 07 · Sep 2025 — Feb 2026

Cook Off 10.0

Scalable Competitive Coding Portal (CodeChef-VIT)

Engineered the real-time competitive coding portal for CodeChef-VIT's flagship contest with 1,000+ active participants. Built a multi-language CodeMirror 6 IDE (supporting C++, Python, Java, and JavaScript) backed by triple-layer state persistence (Zustand in-memory store, LocalStorage write-through cache, and debounced MongoDB sync) and Axios auth/retry interceptors to guarantee zero lost submissions under 40% peak concurrent load.

1,000+ Coders
active contest participants
3-Layer Sync
Zustand · LocalStorage · MongoDB
40% Peak Load
sustained with 0 dropped submits
  • Next.js
  • React
  • TypeScript
  • JavaScript
  • C++
  • Tailwind CSS
  • Zustand
  • CodeMirror 6
  • MongoDB
CODEMIRROR 6 · 3-LAYER PERSISTENCECODEMIRROR 6C++ · PY · JSZUSTAND STOREin-memory stateLOCALSTORAGE0ms recoveryAXIOS PIPELINEJWT · retry queueMONGODB SYNCcloud checkpointJUDGE VERDICT40% peak load
Trace · Contest Submission

Submissions pass from CodeMirror 6 and Zustand through authenticated Axios retry interceptors, checkpoint to MongoDB, and return live judge verdicts under 40% concurrent load.

Timeline

Full-Stack Developer (Team Lead)

Apr 2025 — Oct 2025
Moss X

Architected full-stack e-commerce flows for 100+ users using React, Express.js, and MongoDB aggregation pipelines. Cached the product catalog API in Redis to cut response latency by 90% (1.5s to 150ms) and keep homepage latency under 50ms. Integrated Razorpay payments with HMAC webhook verification and idempotency keys across 20+ live orders, containerized services with Docker, and led 3 developers through PR reviews and delivery.

Read the case study
  • JavaScript
  • Node.js
  • Express.js
  • REST APIs
  • MongoDB
  • Redis
  • React
  • Tailwind CSS
  • Docker

Frontend Engineering Intern

Sep 2024 — Feb 2025
Nepbyte Technology

Built responsive React interfaces with Tailwind CSS, Zustand state management, and Axios REST interceptors. Improved Core Web Vitals through route-based code splitting, lazy asset loading, and semantic HTML — reducing Largest Contentful Paint (LCP) by 49% from 3.5s to 1.8s and raising URLs passing Core Web Vitals from 45% to 85%.

  • React
  • TypeScript
  • JavaScript
  • Tailwind CSS
  • Zustand
  • REST APIs

Technologies

Hover or tap any technology to see the projects and roles where I've used it.

languages

backend

databases

ai

frontend

devops

Get in touch

Hiring for a software, backend, ML, or full-stack role, or building something interesting? I'd like to hear about it.

Email me