AI/ML Engineer (Remote)

Parrative AI, Inc.; United States; Remote; USD 150,000 - 210,000; Full time

Benefits offered by this job

Stock options

Job summary

Parrative AI in a remote USA setting seeks an experienced ML/AI leader to own end‑to‑end LLM pipelines for regulated financial advice workflows. You will build and optimize extraction, reasoning, tool use, and generation, with human‑in‑the‑loop approvals and strong governance.

The role focuses on production systems, evaluation frameworks, and secure, auditable deployments, balancing cost, latency, and reliability in a cloud environment.

Qualifications

• 5+ years shipping ML or applied AI to production, including LLM-based systems.
• Deep fluency in Python and the modern LLM stack: structured outputs, retrieval, function calling, agents, and evals.
• You've built an evaluation framework and used it to stop a launch that wasn't ready.
• Strong engineering fundamentals: APIs, data pipelines, observability, and cloud infrastructure.
• You treat cost and latency as first-class design constraints, not afterthoughts.
• Speech, transcription, or speaker diarization.
• Experience in a regulated industry: finance, healthcare, insurance, or legal.
• Fine-tuning, distillation, or running models in your own infrastructure.
• Shipping inside SOC 2 or similar security programs.

Responsibilities

• Own the end-to-end LLM pipeline: extraction, reasoning, tool use, and generation in production.
• Define evaluation metrics and maintain live production dashboards for model quality.
• Orchestrate agents with human-in-the-loop approvals within the system design.
• Implement retrieval over client history with strict tenant isolation.
• Balance model choices (frontier vs fine-tuned) against cost and latency.
• Ensure governance, security, and auditability across the stack.

Skills

Python

LLM stack

APIs

Data pipelines

Observability

Cloud infrastructure

Cost latency

Speech transcription

Regulated industry

Fine-tuning/distillation

SOC 2

Job description

Build the AI a trillion-dollar industry Needs.

Reports toParrative's CT O

Location: Remote (USA)

Compensation: Competitive base + stock options

The financial advice economy runs on promises. You'll build the intelligence that keeps them.

Every week, financial advisors make thousands of commitments to the households they serve: a document to send, a beneficiary to update, a call before a decision has to be made. Today most of those commitments live in someone's memory. At Parrative, they live in a system that captures them, routes them, and gets them done, with a human approving every step that matters.

You’ll build the intelligence at the center of that system.

The problem

Turn messy, multi-party, high-stakes conversations into completed work. That means:

• • Extraction you can stake a firm's reputation on. Pull commitments, owners, deadlines, and intent out of meetings, emails, and notes with precision a regulated business can rely on.
• • Agents that act, not just suggest. Draft the follow-up, schedule the review, prepare the paperwork, chase the open item, and know when to stop and ask.
• • Every action explainable. What the model saw, what it decided, who approved it, and when. In this industry, AI that can't show its work can't be used.
• • Configurable by policy, not by fork. Each firm shapes how the AI behaves under its own rules, through our patent-pending configurability layer, Fiduciary Forge™.

Why this is the hard version of applied AI

Most AI products can afford to be wrong. A missed suggestion in a consumer app costs nothing. In wealth management, a missed commitment costs a client relationship or draws a regulator's attention. Precision, recall, latency, cost, privacy, and auditability all have to hold at the same time, in production, every day.

If you've been looking for an applied AI problem where getting it right actually matters, this is it.

Engage all three and deals move fast. Cold to sold in a single call is not unusual here.

What you own

• • The LLM pipeline, end to end: extraction, reasoning, tool use, and generation, from architecture to quality in production.
• • The evaluation system: golden datasets, regression suites, human-review loops, and live production metrics. We ship on evidence.
• • Agent orchestration with human-in-the-loop approval built into the design, not bolted on afterward.
• • Retrieval over each firm's client history, with strict tenant isolation.
• • Model strategy: when to use frontier models, when to fine-tune, when to go smaller, and what each choice costs per advisor.
• • Guardrails, PII handling, and the audit trail, built alongside our security and compliance leaders.

What you bring

• • 5+ years shipping ML or applied AI to production, including LLM-based systems that real users depend on.
• • Deep fluency in Python and the modern LLM stack: structured outputs, retrieval, function calling, agents, and evals.
• • You've built an evaluation framework and used it to stop a launch that wasn't ready.
• • Strong engineering fundamentals: APIs, data pipelines, observability, and cloud infrastructure.
• • You treat cost and latency as first-class design constraints, not afterthoughts.
• • Speech, transcription, or speaker diarization.
• • Experience in a regulated industry: finance, healthcare, insurance, or legal.
• • Fine-tuning, distillation, or running models in your own infrastructure.
• • Shipping inside SOC 2 or similar security programs.

What this role is not

• - Not a research seat. Everything you build goes in front of advisors.
• • Not a prompt-tuning job. You own the system, its metrics, and its failure modes.
• • Not a wrapper. The model is the easy part. The hard part is everything around it, and that's the job.