Lead Product Manager

What you'll do

• Own product direction across the full model lifecycle (data, training and adaptation, evaluation, release, production monitoring, and improvement or retirement) for custom SLMs, ASR stack, and real-time inference infrastructure.
• Own the data strategy: acquisition, consent and usage rights, sampling, and annotation.
• Turn ambiguous model-quality questions into measurable decisions and define what "good" means.
• Participate as a peer in eval reviews, error analyses, and incident retrospectives; form independent hypotheses from conversation traces.
• Treat internal teams (voice agents, agentic runtime, AI features, GTM) as customers, providing model capabilities, latency/cost envelopes, and a reliable roadmap.
• Make trade-off decisions on model quality vs. streaming latency, train vs. fine-tune vs. buy, model size vs. capability, and GPU cost vs. price point.
• Write direction memos, decision docs, and specs; make release/rollback calls; own the model roadmap (including deprecations); set data investment priorities; define and measure quality bars.

Skills you'll bring

• Hands-on track record building or running models in production (trained, fine-tuned, served, or optimized); experience with speech/ASR or real-time constraints is highly valued.
• 2+ years of product ownership (formally titled or not), with accountability for what was built and whether it worked.
• Fluency across the model and serving stack: eval design, fine-tune vs. train vs. distill, quantization, serving trade-offs, ASR metrics beyond WER, and real-time inference cost drivers.
• Judgment under uncertainty and direct communication; ability to commit to outcomes and document decisions in writing.

Nice to have

• Speech experience: training or productionizing ASR/TTS, telephony, streaming latency.
• Built training data pipelines or run labeling operations (sourcing, sampling, annotation quality, data rights).
• Run inference infrastructure at scale (GPU capacity planning, serving optimization, cost-per-call tuning).
• Built or run an eval harness in production.
• Experience pricing or packaging AI products.
• Publications, open-source contributions, or technical blog.