Applied Scientist (Multimodal AI)

Invyte.ai; India; On-site; INR 2,500,000 - 4,500,000; Full time

Job summary

Invyte.ai is building advanced AI that understands user context through sensors, memory and device state to drive real-time actions. You will work on multimodal perception, on-device intelligence and agentic systems with tight latency constraints.

The role emphasizes research-driven deployment, model adaptation and structured evaluation to hit product budgets and power limits while maintaining performance.

Qualifications

• Minimum 3+ years of professional experience.
• Experience with agentic systems: planning, decomposition and state management.
• Experience applying quantization, pruning and architecture search for efficiency.
• Applied vision-language models: prompt architecture and visual understanding.

Responsibilities

• Help build an AI system that understands user context from camera, microphones, device state and memory.
• Turn context into useful actions by coordinating with mobile and hardware engineers.
• Contribute to evaluation infrastructure, data pipelines and real-time latency optimization.

Skills

Multimodal AI

Real-time voice

Vision-language models

Memory systems

On-device intelligence

Agentic systems

Reinforcement learning

Model adaptation

Job description

This role sits at the intersection of multimodal AI, real-time voice, vision-language models, mobile-use agents, memory, on-device intelligence, and production-grade evaluation systems. You will help build an AI system that can understand the user’s physical context through camera, microphones, device state and memory - then turn that context into useful action.

• Minimum 3+ years of professional experience.
• Agentic systems - multi-step task execution, planning and decomposition, state management
• Apply quantization (PTQ and QAT), pruning, and architecture search to hit per-product size, latency, and power budgets and efficient in working with SLM’s
• Applied vision-language models - prompt architecture, structured output, context management, and visual understanding of interfaces, design and execute distillation strategies.
• Scale simulation and scaffolding environments for agentic RL: code execution sandboxes, computer use environments, tool-calling harnesses, and Mobile Use Agents.
• Evaluation infrastructure - task suites, automated scoring, regression harnesses, and success / latency / cost measurement
• Data pipelines - capture, schema design, labelling, privacy-safe handling, and dataset curation for training
• Real-time voice - ASR and TTS integration, streaming interaction, and end-to-end latency engineering
• Deep understanding of reinforcement learning: policy optimization, reward design, exploration, and the interplay between environment design and agent behavior.
• Model adaptation - fine-tuning workflows, dataset construction, and running models under tight compute and memory budgets
• Partner with our mobile and hardware engineers to move capability from the cloud onto the device