AI Engineer, LLMs and Autonomous Agents; R&D

Position: AI Engineer, LLMs and Autonomous Agents (R&D

About Ascend

Backed by private equity from people‑focused Alpine Investors, Ascend is building a dynamic platform for regional accounting firms that enables them to stay independent while accessing the resources of a Top 25 firm to help them grow. Ascend provides access to growth capital, robust talent acquisition and training, leading quality management resources, best‑of‑breed technology, AI tools, an offshore captive, shared back‑office services, a catalytic leadership system, and modernized equity incentives so that firms can surmount today’s industry challenges to reach their full potential.

For more information, visit About Alpine Investors

Alpine Investors is a people‑driven private equity firm that is committed to building great companies by working with, learning from, and developing exceptional people. Alpine prides itself on fostering cultures where people value empowerment, diversity, fairness, integrity, and intellectual honesty. Founded in 2001, Alpine specializes in investments in companies in the software and services industries. Alpine has over $15 billion in AUM and has offices in San Francisco, New York, and Salt Lake City.

For more information on the firm, visit Why this role exists

Ascend is pairing deep industry know‑how with cutting‑edge generative AI to reinvent high‐stakes, document‐heavy workflows. You’ll design and ship the autonomous agent framework that makes those workflows vanish behind a chat‐first interface — reducing turnaround from days to minutes and freeing experts to focus on judgement, not data entry.

In this role, you’ll work with an incredible product team dedicated to incredible quality and blazing high product velocity. You’ll work in deep collaboration with our users, our product folks, and our engineers to bring your great ideas to our users’ fingertips in just a matter of hours or days. What You’ll Build • Agentic workflows: Prototype and deploy autonomous agents that take a request, gather context, decide next steps and deliver finished work with minimal human touch. • Agent Tooling & Action Interfaces: Design the tool‑API layer that lets agents trigger real-world actions (emails, filings, database updates) safely and auditable from day one. • Context pipelines: Build the backbone of our agents: a lightning-fast context engine that digests messy inbox threads, docs, and notes, and serves only the gold-standard facts to the model. • Evaluation at scale: Own the whole eval stack, datasets, lightning-fast benches, and live telemetry, so we can prototype in hours, A/B with real metrics, and ship to prod (or roll back) with total confidence. • Data-Privacy Guardrails: Embed red-team tests, PI-scrubbing, and role-based access gates directly into the inference path, so safety is baked in, not bolted on. • Developer Canvas / Playground: Ship an internal playground where PMs and engineers can compose, test, and share agent chains in minutes, no local setup required. In 12 Months, We’ll Measure Your Success By • Agent adoption: Your autonomous workflows and tool interfaces are processing thousands of real requests every week, with low hand-off rates and rave reviews from specialists. • Prototype-to-prod velocity: Idea-to-production time has dropped to single-digit hours, powered by your eval stack, dev canvas and automated rollbacks. • Context engine reach: The knowledge graph indexes millions of secure artifacts (emails, notes, docs) with sub-second retrieval, fueling every agent decision. • Privacy and trust: Red-team drills and external audits pass with zero critical findings thanks to the guardrails you embedded in the inference path. • Team efficiency lift: Internal squads now automate repetitive tasks (test writing, migration scripts, doc drafts) with the playground you shipped, saving 10+ dev-hours per person each month. You’ll Thrive Here If You • Speak the language of product & UX, translating fuzzy user pain into concrete agent capabilities. • Default to automation over manual toil and actively spread AI tooling that makes the whole team faster. • Have shipped production-grade LLM-powered agents and know the nitty-gritty of secure tool invocation, rate limits and hallucination defence. • Can…