Software Engineer - 2 (Voicebot)

4 hours ago

Bengaluru, IndiaMid Level

Responsibilities

Build and maintain LLM and speech evaluation frameworks covering task success, hallucination, instruction following, WER, latency, barge-in, and turn-taking quality.
Run full fine-tuning and PEFT/LoRA/QLoRA experiments on open-weight models and determine when fine-tuning outperforms prompting.
Benchmark LLMs and ASR/TTS engines across cost, latency, and quality, then recommend model and provider choices.
Diagnose and fix production conversation failures involving turn-taking, recognition, latency, and prompt regressions using logs, traces, and evaluation data.
Ship changes to the live conversational pipeline with instrumentation and alerting, and own design, evaluation, deployment, and monitoring.
Collaborate with voicebot architecture, product, and field delivery teams serving enterprise customers.

Strong understanding of artificial neural networks, transformer architectures, attention, tokenization, and decoding strategies.
Hands-on experience building or running LLM evaluation harnesses, LLM-as-judge systems, and regression suites.
Hands-on fine-tuning experience with PEFT, LoRA, or QLoRA on at least one open-weight model for a real task.
Working knowledge of speech and ASR/TTS evaluation, including WER, latency, diarization, and common audio failure modes.
Strong Python skills and ability to write and maintain production code.
2–4 years of software or ML engineering experience, including experience with a production system.
Demonstrated ownership of changes shipped to and operated in production.
Experience with real-time audio or streaming systems, agentic orchestration, LLM tool-calling, RAG, embeddings, vector stores, open-weight model serving, AI observability, multi-tenant SaaS, or voice AI and contact-center domains is preferred.

Work on production-scale AI handling live voice conversations for enterprise customers.
Opportunity to work across the LLM, speech, real-time orchestration, and infrastructure stack.
Collaborate with senior engineers while owning changes through build, evaluation, deployment, and monitoring.

501-1,000 employees