Kapture CX is an AI-native customer experience platform helping enterprises turn their messiest service operations into seamless, intelligent experiences. We build cutting-edge AI (agentic automation, voice AI, real-time intelligence) for industries where complexity is the default: banking, financial services, retail, ecommerce and beyond.

Headquartered in Bangalore with a growing presence in the US and Dubai, we serve 1,000+ enterprise clients across 18+ countries. We’ve raised $20M to date, we’re profitable, and we’re growing 100% year on year, a combination that’s rare at the intersection of AI and SaaS.

Today we’re a 400+ strong team, and we’re looking for people who want to build AI that actually ships, scales and solves real problems.

This is a model-first role where you will research, design, fine-tune, and evaluate AI models powering our conversational call agent platform. You will work across the full model development lifecycle — from understanding problems deeply and generating training data to implementing cutting-edge finetuning techniques and delivering production-ready models.

Here’s a more detailed description of what you will do in this role:

You will research, evaluate, and select foundation models (LLMs, ASR, TTS) based on performance and production constraints.

You will develop end-to-end finetuning pipelines using techniques such as SFT, DPO, LoRA, QLoRA, and other efficient methods.

You will design and implement robust evaluation frameworks including ASR metrics (WER/CER), TTS quality metrics (MOS/UTMOS), and LLM task-specific benchmarks.

You will architect synthetic data generation pipelines to create scalable and high-quality training datasets.

You will collaborate with MLOps teams during model handoff, ensuring clarity on model characteristics, inference requirements, latency, and quantization considerations.

You will adapt models for Indian language and accent diversity, including Hindi, Hinglish, and regional variations.

This is a Bangalore-based role. We work five days a week from the office, as we believe in-person interactions fuel innovation and agility.

You deliver well-characterised, production-ready models that improve conversational intelligence while balancing accuracy, efficiency, and real-world deployment constraints.

Here are the basic requirements:

You have 2–3 years of hands-on experience in ML model development, especially in LLMs, ASR, or TTS.

You are proficient in PyTorch and understand training dynamics, debugging model instability, and optimization techniques.

You have experience working with the HuggingFace ecosystem including Transformers, PEFT, TRL, and Datasets.

You have practical experience with finetuning techniques such as SFT or DPO.

You understand model architecture trade-offs including parameter size, attention mechanisms, KV cache behaviour, and quantization sensitivity.

You can build maintainable, modular Python training pipelines.

You design structured experiments and benchmarking frameworks for model evaluation.

You communicate technical trade-offs clearly to engineering and infrastructure teams.

You have experience finetuning Whisper-family, NeMo, ESPnet, or similar ASR models.

You have worked with neural TTS systems and understand prosody modelling and code-mixed synthesis.

You have experience finetuning open-source LLMs like LLaMA, Mistral, Qwen, or Gemma.

You have exposure to synthetic data generation pipelines or multilingual NLP/speech systems.

You use experiment tracking tools like Weights & Biases or MLflow.

Here’s what you will gain from this role:

Opportunity to work on cutting-edge AI powering real-world conversational systems.

Hands-on exposure to LLM, ASR, and TTS model development at production scale.

Collaboration with highly skilled engineering and AI teams in a fast-growing global SaaS organization.

Strong growth opportunities with competitive compensation and performance-linked rewards.