Explore Veyra as a candidate, investigate the product, evaluate conversational agent behavior, and help us engineer more adaptive, reliable AI interviewers.
This internship operates at the intersection of Agentic AI, LLM applications, AI evaluation, conversational intelligence, and product intelligence.
Your primary responsibility during the selection stage is to use Veyra as a real candidate, attend interviews, and critically analyze the product from end to end.
After selection, your responsibilities will expand based on your demonstrated skill level:
You don't need previous internships, prior AI company experience, open-source pedigree, or published research papers. What matters to us is:
Can you understand a real product and discover where it breaks?
Can you explain why problems happen and who they affect?
Can you think through pragmatic, realistic engineering solutions?
Can you learn new AI concepts rapidly and apply them with precision?
We look for foundational curiosity rather than deep specialization.
Basic conceptual familiarity with:
Basic knowledge of one or more:
Comfort experimenting with ChatGPT, Claude, Gemini, Cursor, or Copilot. What matters is knowing when to use an AI tool, how to verify its output, and how to use it responsibly.
The application itself evaluates your firsthand investigation of Veyra. Follow these 6 steps:
Sign up for Veyra and complete your candidate profile, target roles, and resume details.
Navigate through the dashboard, coding arena, system design canvas, and settings.
Experience at least two full interviews (e.g. Technical, Live Coding, System Design, or Behavioral).
Actively look for UX friction, latency, repetitive questions, missing features, and failure modes.
Document what happened, why it matters, who is affected, and what Veyra should do instead.
Fill out the comprehensive application form and submit your structured product investigation.
We are not asking “Can you find something wrong?” We are asking “Can you understand why something is wrong?”
“The interview page is slow.”
“After the candidate finishes speaking, there is a noticeable delay before the next response begins. This can make the candidate think the microphone or connection has failed. The system should expose clearer processing state feedback and reduce the latency between final STT and TTS generation.”
Why this matters: Weak observations describe superficial frustration without context. Strong observations pinpoint the lifecycle stage (STT → TTS), user psychology (perceived failure), and a concrete architectural fix (intermediate state feedback).
“The AI asks repetitive questions.”
“The interviewer appears to rely too heavily on predefined question progression instead of using the candidate's previous answer as the basis for follow-up questions. This reduces the perception of an adaptive interview and could be improved using conversation memory plus a follow-up decision policy.”
Why this matters: The strong observation identifies the divergence between rigid script progression and dynamic conversation memory, demonstrating an understanding of how autonomous agents maintain context.
Every application is scored holistically across seven core dimensions.
Can the candidate discover meaningful, high-impact issues across the platform?
Can they explain the root problem rather than merely describe superficial symptoms?
Do they understand basic software architecture, APIs, and client-server flow?
Do they understand fundamental concepts around LLMs, agents, context, tools, and evaluation?
Can they propose realistic, technically feasible solutions rather than wishful thinking?
Can they clearly communicate technical findings in structured, precise language?
Basic reasoning, deductive ability, and systematic inquiry.
Performance-based stipend — determined after the evaluation and interview process based on demonstrated technical understanding, problem-solving ability, AI knowledge, communication, and interview performance.
Complete your Veyra interviews and share your insights.