Tanatasa is built for agentic shopping: AI assistants search our catalogue, compare products, build carts and hand checkout back to the shopper. You will own the AI side of that product, from the MCP server and agent API to the LLM pipelines that keep product data accurate, working remotely from anywhere in Asia-Pacific.
Tanatasa (operated by WELL IN STYLE PTE. LTD.) is a Singapore ecommerce company where shoppers and their AI assistants buy everyday essentials: pantry, household, health, personal care, beauty, baby and pet. Shoppers can browse tanatasa.sg themselves or hand a restock list to an AI assistant such as Meta Muse, Instinct, OpenAI Dots or Claude Duet, compare options and approve the order. We are a small team, so everyone ships work customers see within days.
What you will do
- Design and ship tools on our MCP server and agent API so AI assistants (ChatGPT, Claude, Gemini, Meta AI and others) can search, compare, build carts and check out reliably.
- Build LLM pipelines that extract, normalise and enrich product data: pack sizes, ingredients, variants, categories and claims, with human review where it matters.
- Improve catalogue search and ranking for natural-language requests such as restock lists, substitutions and budget limits.
- Create evaluation sets and automated evals that measure whether agents find the right product, price and quantity, and catch regressions before release.
- Instrument agent traffic end to end: tool-call logs, latency, failure modes and conversion, and turn what you learn into product fixes.
- Keep agent-facing documentation, llms.txt and product feeds accurate so assistants describe Tanatasa correctly.
- Work asynchronously with a small team in Singapore, writing clear design notes and pull requests.
What we are looking for
- 3+ years of professional software engineering experience, including at least 1 year building production features with LLMs.
- Strong TypeScript or Python, and comfortable reading and writing both.
- Hands-on experience with LLM tool calling, structured outputs, embeddings or retrieval, and the judgement to know when not to use a model.
- You have built evals or test harnesses for non-deterministic systems and use them to make decisions.
- Solid SQL and HTTP API design.
- Clear written English and a good async working style: you unblock yourself and keep others informed.
- Based in Asia-Pacific with at least 4 hours of overlap with Singapore time (UTC+8).
Nice to have
- Built or contributed to an MCP server, or shipped integrations with ChatGPT apps, Claude, Gemini or similar agent platforms.
- Ecommerce or marketplace experience: catalogues, search relevance, product feeds or checkout.
- Experience with Cloudflare Workers, D1, Vectorize or other edge platforms.
- Open-source work or writing that shows how you think about AI systems.