Skip to content
Search jobs inside your AI agent. Claude, ChatGPT, Claude Code, Codex, Gemini, Cursor and Grok
Connect

Zavo

NewBe an early applicant

SDE3 (Applied AI)

Posted Updated

Description

We are seeking a hands-on, systems-minded SDE 3 to architect and scale the core intelligence layer of our autonomous multi-agent procurement platform.


Enterprise procurement runs on complex, unstructured documents (RFPs, SOWs, MSAs, vendor quotes, compliance certificates) and messy intake requests that must be transformed into deterministic, auditable workflows. In this role, you will go beyond standard frontier API wrappers—owning open-weight model adaptation and serving, multi-modal document intelligence pipelines, autonomous intake structuring, and AI-driven workflow orchestration across our hybrid Python/Go architecture.


The candidate will have responsibilities across the following functions:


Open-Weight Model Adaptation, Fine-Tuning and Serving:

  • Post-Training & Domain Adaptation: Fine-tune and align open-weight LLMs and Vision-Language Models (e. g., Llama, Qwen, Mistral, DeepSeek families) using LoRA/QLoRA, SFT, and preference optimization (DPO/GRPO) for procurement-specific reasoning, structured JSON extraction, and negotiation guardrails.
  • High-Throughput Inference Infrastructure: Deploy, quantize (AWQ, GPTQ, FP8), and optimize open-weight models for low-latency, cost-efficient production serving using engines like vLLM, SGLang, or TensorRT-LLM—supporting both multi-tenant cloud and air-gapped/on-premises enterprise deployments.
  • Model Routing & Distillation: Design intelligent routing architectures that dynamically balance frontier models and distilled task-specific open-weight models based on task complexity, latency budgets, and enterprise data-privacy boundaries.


Document Intelligence and Unstructured Data Extraction:

  • Multi-Modal Document Parsing: Architect resilient document intelligence pipelines capable of ingesting complex enterprise PDFs, scanned contracts, multi-page pricing tables, and redlined Word documents using VLMs, OCR/layout-aware models (e. g., Docling, Marker, Surya), and custom parsing heuristics.
  • Schema-Enforced Extraction & Grounding: Build deterministic extraction engines (Pydantic v2, Instructor, constrained decoding / Outlines) that convert unstructured vendor bids and legal clauses into strictly typed, citation-grounded data objects with zero hallucinated line items.


Intelligent Intake and Agentic Workflow Orchestration:

  • Autonomous Procurement Intake: Design conversational and form-assisted intake agents that disambiguate vague business requests, classify spend categories, enforce policy compliance, and dynamically assemble execution plans.
  • Multi-Agent Graph Execution: Partner with Backend/ Go engineers to build stateful agentic workflows (LangGraph or custom state graphs) that coordinate specialized sub-agents across intake, vendor discovery, quote comparison, and contract risk flagging—exposing typed contracts over gRPC/Protobuf to our Temporal/Go execution backbone.
  • Evaluation & Continuous Improvement: Build automated golden-dataset benchmarks and LLM-as-a-Judge evaluation harnesses to track extraction precision/recall, tool-calling accuracy, and latency regressions across model updates.


Requirements:

  • Experience: 8+ years of software engineering experience, with at least 1+ years building and shipping production LLM, NLP, or multi-modal AI systems at scale.
  • Production Python & ML Ecosystem: Expert-level proficiency in modern Python (3.11+), asyncio, Pydantic v2, PyTorch, Hugging Face (transformers, peft, trl), and high-performance API frameworks (FastAPI).
  • Open-Weight Training & Serving: Hands-on experience fine-tuning open-weight models (LoRA/QLoRA, synthetic data generation) and serving them in production via vLLM, SGLang, or Triton/TensorRT-LLM on GPU clusters (AWS/GCP).
  • Document Intelligence & Retrieval: Proven track record extracting structured data from complex documents (tables, layouts, OCR, and vision-language models) and building hybrid retrieval pipelines (dense/sparse embeddings, rerankers, and vector stores).
  • Agentic Systems & Structured Outputs: Deep understanding of tool-calling architectures, constrained grammar decoding, multi-step planning loops, and systematic evaluation methodologies.
  • Systems and Engineering Rigor: Strong software engineering fundamentals—able to profile GPU/CPU memory bottlenecks, design clean microservice boundaries, and collaborate on gRPC/Protobuf interfaces with distributed systems engineers.


Preferred Qualifications

  • Prior experience deploying self-hosted / on-premises AI stacks for security-sensitive enterprise customers (SOC 2, GDPR, or air-gapped environments).
  • Domain exposure to enterprise procurement, contract lifecycle management (CLM), legal tech, or financial document processing.

Skills

What Senior AI Engineering jobs ask for — and how much of it you have →
Apply

See also

AI Engineering jobs by country — openings, pay and top skills →