Position: AI Research Engineer – Small Language Models & On -Device AI
Work Type: Full Time/Hybrid (India)
Company Location: Surat, Gujarat

At VideoSDK, we’re building the real -time intelligence layer for the next generation of applications — quick, private, multimodal, and on -device.

We're hiring an AI Research Engineer to lead infrastructure development for small language models (SLMs) and speech systems, optimized for on -device multimodal AI. Your work will directly power products like real -time voice agents, live translation, intelligent video systems, and low -latency assistants — all running beyond the cloud.

What You'll Work On


- Build scalable infrastructure to train, evaluate, and deploy small language models efficiently.

- Design and implement RL -based training loops (e.g., PPO, DPO, RLAIF) for tuning small models in constrained environments.

- Work with speech systems, including speech -to -text, text -to -speech, and voice activity detection.

- Optimize models for on -device inference – targeting mobile, browser, and edge hardware (CPU, GPU).

- Contribute to building real -time multimodal AI pipelines combining text, audio, and video.

- Translate cutting -edge research into clean, production -ready code.

- Drive experiments on quantization, distillation, and architecture search for efficient deployment.

What We're Looking For

- Strong understanding of training, testing,



and fine -tuning small models (≤2B parameters).

- Experience with reinforcement learning for language models (GRPO, PPO).

- Proven experience working with speech models (Whisper, xTTS, Silero etc).

- Ability to read and implement research papers quickly and effectively.

- Solid Python and PyTorch skills with an eye for clean, modular code.

- Experience with building and managing custom datasets, training pipelines, and evaluation suites.

- Familiarity with multimodal AI, particularly combining audio, text, or video.

Bonus Points

- Experience optimizing inference for mobile or edge deployments (ONNX, CoreML, TFLite, MLX, WebAssembly).

- Knowledge of state space models, recurrent architectures, or other attention -free designs.

- Contributions to open -source AI projects.

- Familiarity with latency -constrained or real -time systems.

Why Join VideoSDK?


We’re pushing the boundaries of real -time, personalized, multimodal AI — and doing it on the edge. You’ll work in a fast -moving environment where research meets product, and where your work directly shapes next -gen communication, assistants, and live experiences.

This is your chance to build the infrastructure behind real -time, human -like AI systems that don’t just live in the cloud — they live everywhere.