Pradyuman Sharma
Software Engineer
pradyumanmusic@gmail.com+91 85808 48682Himachal Pradesh, Indiaportfolio-pradyuman.netlify.appurnextdoor.com/users/pradyumanmusicgithub.com/PradSharma554
Summary
Software engineer working across TypeScript, Node.js, Next.js and PostgreSQL. Spent a year at zop.dev building Zopnight, a multi-cloud cost optimization platform managing 5k+ cloud resources. Independently built CloudAIr, an access layer between websites and AI agents, and LLM Execution Inspector, a tracing and observability tool for LLM applications. Looking for full-stack and backend engineering roles.
Experience 1 yr
- Built core features for Zopnight, a multi-cloud (AWS/GCP/Azure) cost optimization platform managing 5k+ cloud resources: scheduling, budgets, cost recommendations and unified dashboards
- Built a virtualized table supporting 50k+ rows with filtering, column reordering and nested row expansion
- Implemented API-driven state management with TanStack Query and Zustand, reducing redundant API calls by 40%; integrated PostHog and Bugsnag
- Raised Lighthouse scores from the 60-80 range to 100 across Performance, Accessibility, Best Practices and SEO through code-splitting, lazy loading and bundle optimization
- Set up a Storybook-driven atomic design component architecture
Skills: TypeScript, JavaScript, React.js, Next.js, Tailwind CSS, Git, GitHub, CI/CD, Docker, Node.js, REST APIs
Education
Skills
TypeScript, JavaScript, Node.js, Express.js, Fastify, PostgreSQL, Redis, REST APIs, React.js, Next.js, Tailwind CSS, SQL, C++, Docker, Git, GitHub, CI/CD
Projects
- Access layer between websites and AI agents (GPTBot, ClaudeBot, Perplexity). TypeScript, Express, Next.js, PostgreSQL, Gemini API, MCP. Apr 2026 - Jul 2026.
- robots.txt-compliant crawler generates each site's llms.txt via Gemini with content-hash caching and a non-LLM fallback; serves clean markdown mirrors and exposes site content through per-site Model Context Protocol (MCP) endpoints
- Crash-resumable background crawls on ephemeral free-tier infrastructure using per-page PostgreSQL checkpoints with heartbeat-based recovery, verified by resuming a SIGKILL-interrupted crawl mid-run with zero lost work
- Per-agent token-bucket rate limiting with owner-configurable rules, identification of 30+ AI crawlers via a signature table, and a non-blocking buffered analytics pipeline with batch inserts and daily rollups; deployed on Render/Vercel/Neon at $0/month
- Chrome DevTools for AI applications: LLM tracing and observability. TypeScript, Fastify, PostgreSQL, Cloudflare R2. Aug 2026 - Present.
- Tracing SDK wraps any LLM client via a JS Proxy, using AsyncLocalStorage to reconstruct execution trees (retrieval, tool calls, retries) with zero context threading
- Captures time-to-first-token and prompt-cache token splits from the SSE stream; cache-aware pricing corrected cost 6.5x versus naive input-token math ($0.081 vs $0.527 per trace)
- Cut trace payloads 250x (72KB to 290B) with SHA-256 content-addressed storage and async gzip, keeping Neon's 0.5GB free tier viable; hardened the public collector with per-key rate limiting and byte-based storage quotas
Certifications
- Elite Silver Badge - Cloud Computing (76%), NPTEL
Honors & Awards
Four merged pull requests.