About the role
Taskly is hiring an AI Engineer (you may know the job as Generative AI Engineer or LLM Engineer) to own the language-model features inside the product. A poster types, speaks or photographs a task. Your code turns that into a category, a moderated post and a price suggestion. These features run on Groq-hosted Llama and GPT-OSS models today, and you will make them more accurate, faster and cheaper.
You will work directly with the founder and the engineers who build the posting flow and the offer screens. The team is small and remote across India, the US, Canada and the UAE, with a shared window of 8:30–10:30 am Eastern. Six months in, we want the classifier measured against an eval set you built, a Zen support assistant that answers real questions correctly, and a cost-per-request number you can explain to anyone.
What you’ll do
- Improve the classifier that sorts any task into a taxonomy of about 1,300 subcategories. Start with the confusing edge cases, where a wrong pick means a poster sees the wrong Taskers.
- Build eval sets for classification, moderation and price suggestions. Run them before every prompt or model change ships, so a regression shows up as a failed check.
- Ship RAG features on Postgres with pgvector, including semantic search across tasks and the grounded answers behind the Zen support assistant.
- Extend voice and photo posting so a task spoken in Hindi or photographed on a phone arrives as a clean, correctly categorized post with a sensible price range.
- Cut latency and cost per call with caching, smaller models, routing and structured outputs. Report both numbers to the team every week.
- Write the guardrails: prompt-injection handling, moderation checks and graceful fallbacks. Posting should still work when a model provider is slow or down.
- Work with the founder to decide when a feature needs a model at all. Sometimes a plain rule is cheaper, faster and easier to explain.
What you’ll need
- 2+ years building and shipping LLM-powered features to real users, in Python or TypeScript, with at least one still running in production.
- Hands-on experience with prompt engineering, structured outputs and tool calling through the OpenAI API, Groq or a similar provider.
- You have built an eval harness and used its results to choose between two prompts or two models.
- Working knowledge of RAG and vector search: embeddings, chunking, reranking and a store such as pgvector.
- Comfortable in SQL, and able to read production logs to find out why a model call failed, timed out or cost too much.
- Clear written English. You can explain a trade-off between quality, latency and cost in a note the founder can decide from.
Education: A degree in computer science, statistics or a related field is welcome but not required; LLM work you have shipped and measured matters more.
Nice to have
- Experience with LangChain, LlamaIndex or a hand-written orchestration layer, and a clear view on when not to use one.
- Multilingual NLP work in English, Hindi or Hinglish, including how models handle mixed scripts.
- Experience at a marketplace (Airbnb, Uber, Urban Company, Swiggy, Airtasker, TaskRabbit, Upwork) or another two-sided product.
- Fine-tuning or distilling a small model for one narrow task, and knowing when it was worth the effort.
Skills and tools
- Python
- TypeScript
- LLMs
- Generative AI
- RAG
- Prompt Engineering
- LLM Evals
- pgvector
- Vector Search
- Groq
- OpenAI API
- LangChain
- PostgreSQL
Where you’ll work
- Remote. Fully remote. Open to people who live and can legally work in India, the United States, Canada or the United Arab Emirates. You work from your own country, so there is no relocation and no visa to arrange.
- Shared hours. The team overlaps for two hours each weekday: 8:30–10:30 am Eastern · afternoon in Dubai, evening in India. The rest of your day is yours to arrange.
How we hire
- 1Apply in about five minutes: your resume and a few questions
- 2A 45-minute conversation with the person you’d work beside
- 3One paid day on a real problem
- 4A decision in writing, with the reason
A person replies to every application within ten working days, even if it’s a no.
About Taskly
Taskly is a marketplace for getting anything done. Someone posts a task in their own words — typed, spoken or as a photo — and people who have been reviewed and approved send offers at a firm price. They pick one, and it gets done: a TV mounted, a website built, a parent taught to use their phone. We’re live in Canada and India and opening to more countries, built by a small team that works remotely across India, the US, Canada and the UAE.
Questions about this role
Straight answers to what people ask most.
Is the AI Engineer role remote?
Yes, fully remote. You can work from India, the United States, Canada or the United Arab Emirates, and the team overlaps for two hours each weekday: 8:30–10:30 am Eastern · afternoon in Dubai, evening in India.
How much experience do I need?
About 2–6 years. It’s a guide, not a cut-off: if you can do the work described here, apply.
What does the AI Engineer role pay?
Pay isn’t listed here yet. It depends on the country you work from and your experience, and it’s discussed openly in your first conversation.
How do I apply?
Upload your resume on the application page and answer a few short questions. It takes about five minutes, with no cover letter and no account. A person replies within ten working days, yes or no.
What does the hiring process look like?
A 45-minute conversation with the person you’d work beside, then one paid day on a real problem, then a decision in writing, with the reason.
Do I need a machine learning degree to apply?
No. We care that you have shipped LLM features to real users and can show how you measured whether they worked. A degree is welcome but not required.
Which models will I work with?
Today the product runs on Groq-hosted Llama and GPT-OSS models. You are free to compare other providers, and we choose by eval results, not by habit.
Something else? Ask it in your application. A person reads it and answers within ten working days.
We don’t use AI to screen, score or rank applications: a person reads every one. If you upload a resume, AI reads it only to pre-fill your form, and you check every field before you send.
Taskly is an equal opportunity employer. We welcome people of every background, gender, age, religion, disability, sexual orientation and nationality, and we don’t ask about any of them. If anything about the process doesn’t work for you, tell us in your application and we’ll change it.
Apply for AI Engineer
Your resume and a few questions. About five minutes, no cover letter, no account.
Apply now →