Artificial Intelligence Engineer

Build the AI that teaches. Our companion doesn't wait to be asked — it decides what the student does next, explains when they're stuck and knows when they're ready to move on. Behind one personality sit specialist agents for explanation, memorization, and practice, plus a router that picks the right one. That system is yours.

KEY RESPONSIBILITIES

• Build the agent architecture: specialist agents for explanation, memorization, and practice behind a single companion
• persona
• Build the routing layer that knows the student's context and calls the right specialist at the right time
• Build RAG over NCERT and curriculum content, tied to the concept hierarchy
• Make the companion teach well in Hinglish — tone, empathy, and code-switching that sounds like a real teacher
• Own evaluation: is the explanation correct, is it pitched right, did the student actually learn — automated and human-in-the-loop
• Implement mastery scoring and the logic that decides what a student sees next
• Own guardrails and safety for a student-facing AI, and keep latency and cost in check at scale

REQUIRED SKILLS

• 5+ years in applied ML or NLP
• Strong Python (PyTorch / TensorFlow / scikit-learn)
• Hands-on with LLMs: prompt engineering, RAG, embeddings, vector stores
• Built multi-agent or tool-using systems in production, not just demos
• Ability to design an eval suite, not just run one
• Cloud ML infrastructure (GCP Vertex AI preferred; SageMaker / Azure ML also fine)
• Experience taking ML systems from prototype to production

NICE TO HAVE

• Education or tutoring products — knowledge tracing, mastery models, spaced repetition
• Indian language NLP and Hinglish code-switching
• Production experience with Gemini / Claude / GPT-class models
• Speech and voice AI (Whisper, Deepgram, TTS)
• Serving LLMs cheaply at scale — caching, routing, smaller models where they suffice

HOW WE WORK

• Small teams with real ownership — you shape how your area works, rather than inheriting a process
• We ship fast, measure honestly, and cut what isn't working instead of defending it
• Research and data sit next to product here; opinions lose to evidence
• AI tooling is part of the workflow for everyone, not just the engineer