Data Scientist, Dusty

• Experience: 2+ years building production ML/NLP or agentic systems; capable of owning projects end-to-end as a senior IC
• LLM/agent skills: Strong with prompt engineering, function-calling/tool use, retrieval (vector DBs + metadata filters), and evaluation harnesses
• Software engineering: Proficiency in Python/TypeScript; shipping cloud services (REST/gRPC), queues, and data pipelines; excellent testing and observability
• Data & search: Hands-on with embeddings, hybrid search, chunking strategies, and latency/recall tradeoffs at scale
• Product mindset: Comfort turning ambiguous field problems into shipped workflows; clear written communication of trade-offs and assumptions
• Bonus: BIM/VDC or construction data familiarity; vision models (OCR, plan understanding); on-device/edge inference; privacy/compliance experience

Desirable

• IC role owning the design, implementation, and iteration of Dusty’s AI agent(s) for construction
• You’ll turn messy, real-world data (BIM, drawings, photos, logs, portal data) into reliable actions and insights, integrating tightly with our products and customer tools
• You'll be joining as one of the early members of a team tasked with building out Dusty Robotic's vision of simplifying construction connecting workers with the information they need to build and lower the cost and improve the quality of the spaces in which we work and play
• Agent architecture & orchestration: Design multi-tool agents (planning, function-calling, tool selection, state/trace) that can reason over BIM/VDC artifacts, field telemetry, and documents
• RAG & retrieval: Stand up retrieval pipelines (chunking, hybrid search, filters, caching) over drawings, specs, RFIs, submittals, and jobsite logs; own evaluation (recall@k, answer quality) and latency budgets
• Tool integrations: Connect agents to internal and partner systems (e.g., design files, project management, issue tracking), with robust auth, rate-limit handling, and audit logs
• Workflow automation: Ship agent skills that draft submittals/RFIs, generate print/verification checklists, summarize shift reports, and surface layout conflicts
• Safety & reliability: Implement guardrails, fallbacks, and red-team tests; define human-in-the-loop paths for high-impact actions
• Data & MLOps: Own prompt/versioning strategy, datasets, offline/edge considerations, telemetry dashboards, and cost controls; partner with infra to productionize models
• Field feedback loop: Work shoulder-to-shoulder with Portal/iPad/robot teams and visit jobsites to iterate on real user needs; document decisions and APIs clearly
• Outcomes (First 6–12 Months)
• Ship a v1 agent that reliably answers and automates a prioritized set of construction tasks with measurable quality and latency SLAs
• Establish an evaluation suite and dashboards for grounding, safety, and cost/1K tasks
• Integrate with at least two high-leverage tools/workflows used by our field and customer teams
• Create internal docs and playbooks so others can extend agent skills safely
• (We use explicit, time-bound outcomes in Dusty scorecards to align expectations.)