AI Engineer (Lead/Principal IC) - Agents, Agentic ...

?? Original Job Listing (use link to apply)is Spindle AI?On a lifelong mission to build software that augments human cognition, Spindle is currently teaching AI agents to help teams see & seize opportunities in their financial plans.We believe that by equipping LLMs to reliably manipulate multidimensional data and numbers (like those found in enterprise financial planning models), we can empower the C-Suite to confidently answer 100x more "What If" questions than they can today. 60-second blurb: started with a"business analytics co-pilot for Strategic Finance" that automatically explores rich "scenario landscapes" to model financial/resource tradeoffs. 6-minute demo video: spanning use cases like pricing & packaging, margin optimization, and sales mix shifts, Spindle AI's private beta has been battle-tested by industry leaders like , IBM Apptio, NewsCorp, and a growing community of enterprises worldwide. Case studies: & TeamBacked by >$6.5 million in disclosed seed funding from Accel Partners (Atlassian, Qualtrics, UiPath, Slack), Sequoia Capital (Snowflake, Clari, Stripe, Okta)Caffeinated Capital (Airtable, Notion, Brex, RevOps), and relevant Fortune 100 executives including the former CFO of Goldman Sachs, CBO of Salesforce, and moreLaunching publicly in 2025 (demo), Spindle was founded by repeat technical co-founders Ryan Atallah (former CTO of BI engine ClearGraph, acquired by Tableau)and Carson Kahn (former President of CIA-backed analytics platform Brainspace AI, acquired by Reveal) with an advisory board from Salesforce, Airtable, Anaplan, Adaptive, and beyondWorld-class, multi-talented product team of 9 (and growing) has previously built and contributed to the leading edge of AI products for data applications & business analytics, including GPT-2/-3, Tableau AI, Precog, Causal, and more. Represents 10+ languages, multiple ethnicities & genders, and uniquely deep cross-disciplinary skill setsLife at Spindle AIAs a remote-friendly company with hubs in San Francisco & Palo Alto, we approach most decisions (including hiring/firing decisions) through six operating principles that make Spindle AI the most fulfilling workplace we've ever built:Invest in relationships.Distill.Always be learning.With bravery; without bravado. ??Choose trust.Cultivate self-knowledge.Because we care deeply about each other's wellbeing, we observe a strict 4-day workweek one week every month, and a 2-week winter holiday company-wide. A few times each year, we also embark on (U.S.-based) destination retreats to explore, grow, and connect more deeply.??? Tech StackOur codebase "” which embraces many of the best practices we all wish we'd had at prior companies "” today consists of ~200k-300k lines of code in:Backend word cloud: Rust (??), Golang, TypeScript, Python & Cython, LLMs, GraphQL, Postgres, DuckDB (??), JTD-CodegenFrontend word cloud: TypeScript/React, WASM, Advanced Browser APIs, Sass, Websockets, GraphQL, JTDInfrastructure word cloud: Terraform, Docker, AWS, Postgres, GitHub Actions CI/CD, SentryYour New RoleIn your mission-critical role as the 10th and final member of Spindle AI's founding team, you'll serve as a hands-on, senior engineer (not data scientist) who hits the ground running to (in the first 8 months)Be the principal implementer and lead architect for Spindle's existing multi-turn, multi-agent system suede_agents shipping performant and magical applied LLM experiences on top of high-leverage, stage-appropriatetechnical foundations.Rapidly advance the agents' analytical reasoning & planning, tool usage, memory & retrieval, and fault recovery with a combination of lightweight & adaptable but principled architectural design, applied science, prototyping/experimentation, and rigorous, production-ready software engineering.Own evals, and a framework for evals(both deterministic and LLM-evaluated) and observability for reliabilitycomplianceperformance, and drift monitoring. Systematize prompt engineering and prompt optimization experiments.Co-lead Spindle's Applied AI Research Program "Agent Smith"closely monitoring the forefront of applied research in LLMs/agentsimplementing new experiments monthly/quarterly with talented colleagues, and helping productionize what works.Alongside the AI for Algorithmic Computing Team, become one of Spindle's 2-3 most frequent inventors and integrators of new tools for the agent, including retrieval, analytical/computational, and reporting tools.Integrate these tools for suede's multi-agent system (Python) with backends for representational computing kevlar (Rust), algorithmic computing suede_algos (Cython), application APIs nylon (Go), and performance-focused frontend for multidimensional data analytics silk (Typescript/React).Propose and test improvements to an AI-native data model for multidimensional data analytics, in close collaboration with the Mathematical Computing Team.Help grow Spindle's core team by proactively attracting, interviewing, closing, and nurturing other world-class team members.Other needs or initiatives that you identify!Hypothetical projects (6-9 months)Lightweight & adaptable but principled rails for multi-agent orchestration, state management, branching, and reflective programmingPrompt component management for the agentic system, later including your own evals & synthetic datasets to test automated prompt optimization or prompt-as-program techniquesNew agent tools for data analytics, mathematical computing, and information retrieval workflows, alongside integrating Spindle's tasks-specific LLM tools and judges into the multi-agent paradigm (if/where reasonable)Capabilities to automatically forecast, rate/rank, and symbolically formulate new analytical, financial, or mathematical questions worth answering for a given business based on its relational  nonrelational data, recent earnings call transcripts, etc.Fine-tuning LLMstuning vector databases for RAG, etc. (but not training or quantizing foundation modelsAutomated multi-LLM/cross-LLM testing & routing, together with other infrastructural experiments for cost controlperformance, and scalingGenerative interfaces that support streaming data, state synchronization?? QualificationsWe view this section as a "guide"not a definitive or firm list of requirements. Even if you don't match everything listed, we encourage you to introduce yourself to us: Spindle AI hires people, not roles; and we hire for strengths, not for lack of weakness. As a team of uniquely diverse backgrounds, ethnicities, ages, genders, and life experiences ourselves - we're also committed to being an equal-opportunity employer. This said, we'll likely interview you about:Skills ???Exceptional command of the architecture, development, testing, and research of LLM-based agents, including both current-generation deployments and next-generation patterns/researchDemonstrableexpertise in a variety of agentic systems techniques, including at least several of (e.g.)fine-tuning, performance tuning of vector DBs, semantic cacheing, tabular RAG/HyDE, dynamic output structuring, concurrent orchestration, MIPRO or automated prompt optimization, program-aided LMs, advanced LLMOps, or othersAs swift at experimentation & prototyping as you are rigorous at integrating & productionizing (this "engineer-scientist" role is a blend of each)Strong mathematical aptitude in any discipline relevant to Spindle, such as (e.g.) optimization, numerical linear algebra, simulation, operations research, or othersComfortable building in compiled, concurrent, type-safe backends for modern applicationsProficient in applying data science or ML methods to both structured data and unstructured data (non-audiovisual)Proactively advocates changes (including changes to this skills list or job description)Strong product sensibilities; aptitude for mathematical thinkingTop-notch written communicationBonus Points ?Demonstrable application of LLMs/agents to business analytics or related fields (such as data analytics, business intelligence, finance, or programming/IDEs); or of tools for thought, explorable explanationsPassion-project experiments with relevant libraries such as (e.g.) DSPy, OpenAI Swarm, SAMMO, LangGraph, Autogen, Qdrant/Chroma/Pinecone, OpenLLMetry, etc.Recent work in our tech stack (described earlier)Substantial passion for a subfield of advanced mathematicsFull-stack or front-end experience building modern web applicationsExpertise or interest in distributed systems and reactive systems; expertise or interest in declarative & functional programmingAvid reader of or commenter on technical writing by the likes of (e.g.) Lilian Weng, Eugene Yan, Chip Huyen, Maxime Labonne, Karpathy, Simon Wilson, swyx, etc. in addition to arXivPrior startup experienceBased in San Francisco Bay Area;or willing to travel here every-other month(ideally based in North America)?? Compensation & Benefits"º Details at Interview ProjectDepending on your background and prior work, we may or may not ask you to complete all or all of a short project as part please of your interview cycle. In either case, ensuring you're comfortable with the kind of project we might ask you to undertake is a useful lens for evaluating your own fit for this role! So, please consider reviewing this example project: To ApplyPlease do not apply [only] via LinkedIn; please apply using the instructions at the bottom of:

Created: 2025-01-28