Full stack and agentic AI engineer, 8+ years engineering (5 full stack, 3 backend), based in India. I build and operate products end to end rather than handing them off.
Shipped 0 to 1, solo or leading a small team:
- Semantic Pen (CTO): AI content platform, 18,000 users, 10,000+ articles generated. Built the RAG pipeline with vector search and an MCP server ecosystem exposing our tooling inside Claude and Cursor. Led 3 engineers. Paddle billing with webhooks and usage-based metering. AWS across Lambda, ECS, S3, SES.
- VerifyForge: email validation API with real-time SMTP verification at 99%+ accuracy, usage metering and Stripe billing. Next.js, TypeScript, tRPC, Postgres, Redis, BullMQ.
- Job aggregation platform: 5,000+ listings, 7 specialized BullMQ queues at 10,000+ jobs daily with retry and dead-letter handling, Scrapy scrapers on AWS Lambda, multi-tenant OAuth.
- Django medical imaging platform: Celery workers for async DICOM parsing, OpenAI function calling through Pydantic-validated tool schemas, 85% pytest coverage.
Other things I have shipped that are easy to verify:
- 2 PRs merged into openclaw (389k stars), the open-source AI agent platform: Discord forum applied_tags support, and a Slack session key fix
- LLM benchmark harness scoring Claude Sonnet 4.6, Haiku 4.5 and GPT-4o-mini on accuracy, latency and cost per correct answer: github.com/pushkarsingh32/ds-portfolio
- MCP server published on npm with a one-click Cursor install: github.com/pushkarsingh32/semantic-pen-mcp-server
- Client SDKs for the same API in 8 languages, published across npm, PyPI, crates.io, NuGet and Packagist
- Open source: 2 PRs merged into openclaw (389k stars), the open-source AI agent platform, both platform integration fixes (Discord forum applied_tags, Slack session key parsing). Also build LLM evals: benchmark harness scoring Sonnet 4.6, Haiku 4.5 and GPT-4o-mini on accuracy, latency and cost per correct answer at github.com/pushkarsingh32/ds-portfolio
Performance work I am happy to talk through: Redis caching taking response times from 2s to 300ms and cutting upstream API spend 60%, and PostgreSQL tuning from 1.2s to 180ms across 25+ composite indexes.
I work through Claude Code and Cursor daily, and I build on these models as well as with them. Most interested in agentic AI, developer tooling, and product engineering roles where I own things end to end.
Remote: Yes, remote only Willing to relocate: Yes
Technologies: TypeScript, Next.js, React, Node.js, tRPC, Python (FastAPI, Django, Celery, Scrapy), PostgreSQL, Redis, BullMQ, Drizzle, Docker, AWS (Lambda, ECS, S3, SES), Cloudflare Workers, LLM APIs (Claude, GPT), RAG, vector search, MCP servers, Stripe/Paddle
Résumé/CV: https://pushkarkathayat.com
GitHub: https://github.com/pushkarsingh32
LinkedIn: https://www.linkedin.com/in/pushkarsingh32/
Email: pushkarsingh32@yahoo.in
Full stack and agentic AI engineer, 8+ years engineering (5 full stack, 3 backend), based in India. I build and operate products end to end rather than handing them off.
Shipped 0 to 1, solo or leading a small team:
- Semantic Pen (CTO): AI content platform, 18,000 users, 10,000+ articles generated. Built the RAG pipeline with vector search and an MCP server ecosystem exposing our tooling inside Claude and Cursor. Led 3 engineers. Paddle billing with webhooks and usage-based metering. AWS across Lambda, ECS, S3, SES.
- VerifyForge: email validation API with real-time SMTP verification at 99%+ accuracy, usage metering and Stripe billing. Next.js, TypeScript, tRPC, Postgres, Redis, BullMQ. - Job aggregation platform: 5,000+ listings, 7 specialized BullMQ queues at 10,000+ jobs daily with retry and dead-letter handling, Scrapy scrapers on AWS Lambda, multi-tenant OAuth.
- Django medical imaging platform: Celery workers for async DICOM parsing, OpenAI function calling through Pydantic-validated tool schemas, 85% pytest coverage.
Other things I have shipped that are easy to verify: - 2 PRs merged into openclaw (389k stars), the open-source AI agent platform: Discord forum applied_tags support, and a Slack session key fix
- LLM benchmark harness scoring Claude Sonnet 4.6, Haiku 4.5 and GPT-4o-mini on accuracy, latency and cost per correct answer: github.com/pushkarsingh32/ds-portfolio
- MCP server published on npm with a one-click Cursor install: github.com/pushkarsingh32/semantic-pen-mcp-server - Client SDKs for the same API in 8 languages, published across npm, PyPI, crates.io, NuGet and Packagist
- Open source: 2 PRs merged into openclaw (389k stars), the open-source AI agent platform, both platform integration fixes (Discord forum applied_tags, Slack session key parsing). Also build LLM evals: benchmark harness scoring Sonnet 4.6, Haiku 4.5 and GPT-4o-mini on accuracy, latency and cost per correct answer at github.com/pushkarsingh32/ds-portfolio
Performance work I am happy to talk through: Redis caching taking response times from 2s to 300ms and cutting upstream API spend 60%, and PostgreSQL tuning from 1.2s to 180ms across 25+ composite indexes.
I work through Claude Code and Cursor daily, and I build on these models as well as with them. Most interested in agentic AI, developer tooling, and product engineering roles where I own things end to end.