AI full-stack and AI automation engineer
Ayush Sinha
I build AI systems that run on real business data, along with the backends and dashboards around them. I live in Mumbai and work US, UK, EU or NZ hours. I'm looking for a long-term remote role on a team, and happy to start with one scoped project.
I have nearly 2 years of paid experience. Most recently I worked on an AWS and TypeScript platform for New Zealand quick-service restaurant franchises. Before that I spent over a year at an AI automation agency, building LLM products for Irish and UK clients in insurance, financial advice, sales and health.
I build with Claude Code agents. Each piece of work starts as a spec, becomes a plan, and ships with tests, and I review every change before it merges. Where a model shouldn't be trusted, such as checksums, money or database writes, I put plain code in charge.
Case studies
Document AI for an Irish life-insurance brokerage
Signed policy documents go through OCR, GPT extraction, validation agents and plain-code checks before anyone sends them to an insurer.
3,205 policies tracked in 7 months
Sales ingest for 180 fast-food stores in New Zealand
A synchronous ingest path rebuilt on SQS FIFO and Fargate workers, then load-tested until it held.
p99 accept latency 13.3s to 1.63s
Talk to your ERP
Plain-English questions over a 64k-order ERP mirror. The model writes a whitelisted query plan that the server validates and runs read-only.
70 of 74 questions answered in the eval sweep
More work
Call-QA scoring, an AI sales platform, a WhatsApp nutrition coach, n8n agent workflows and my delivery record on my last contract.
306 merged PRs in 4 months
Work with me
I offer four fixed-scope engagements: an AI document pipeline, an LLM eval harness, an n8n to TypeScript migration, and an embedded engineer retainer. I work hourly or quote a fixed price per scope. I can also work under an agency's brand.
Stack
Languages
LLM and AI
Backend and frontend
Data and cloud
Testing and AI dev tools
Selected projects
- Fix for a hang in dap, an open-source Go debugger CLI that AI coding agents use. js-debug's child session was never read, so every Node.js debug command hung. My fix comes with a fake-adapter unit test that reproduces the hang in CI, and the project's own Node end-to-end test went from timing out to passing in 3.5s. The PR is open upstream.
- A local semantic code-search setup for coding agents. I set up an existing MCP server (claude-context) with Ollama embeddings and a Milvus vector database, so agents could search 6 proprietary repositories without any source code leaving the machine.
- A parallel-worktree tool in Bash (18 commands) that gives each AI-agent branch its own git worktree, ports and Postgres database, so up to 7 feature branches can be built and tested at once.