Prompt Guide
Learn how to write high-signal search prompts.
A strong search prompt helps DINQ understand who you want to find and what evidence matters most. You do not need to include every detail upfront, but adding the right context can improve the quality of your results.
Blueprint for a high-signal search prompt
A useful search prompt usually includes three types of context:
| Information layer | Example | Why it helps | What to include |
|---|---|---|---|
| Persona | Multimodal model researcher | Anchors the core candidate profile. | Role, title, seniority, and domain focus |
| Evidence | CVPR or NeurIPS papers, open-source projects | Tells DINQ what proof of relevance to look for. | Papers, venues, repositories, shipped products, OSS impact, benchmarks |
| Constraints | North America, remote, senior | Narrows the candidate pool to practical requirements. | Location, timezone, work mode, years of experience, compensation range, education |
Example prompts
Short prompt:
Find AI infrastructure engineers with open-source contributions.
More detailed prompt:
Find senior infrastructure engineers in North America who have worked on LLM inference optimization. Prioritize people with CUDA, TensorRT, vLLM, or vector database experience, and look for strong open-source contributions or shipped AI infrastructure projects.
Research-focused prompt:
Find multimodal model researchers with CVPR, ICCV, NeurIPS, or ICLR publications. Prioritize candidates with recent papers, strong citation signals, and public code or project pages.
Reduce follow-up questions with clear criteria
If your initial prompt already includes the important criteria, DINQ does not need to ask for those details again. You can still review and edit the Search Plan before clicking Start Search; filling in a fixed set of fields does not automatically launch the search.
Helpful criteria include:
- Location or timezone.
- Target title or role.
- Expected work experience.
- Required education level.
- Compensation range, if relevant.
- Must-have or must-exclude conditions.
You can still refine the search later if the first results are too broad, too narrow, too senior, or too junior.