Software Engineer, AI

Vectrel builds custom AI integrations into clients' existing business infrastructure. No templates, no rigid packages. We scope solutions after discovery, deliver in phases, and work directly with the people who own the problem. Our clients span legal, healthcare, insurance, retail, and technology.

This is a small team. You will work directly with the engineering lead on production client work from day one.

We're looking for a software engineer with hands-on AI engineering experience to take on scoped deliverables within active client projects. You'll build backend systems, integrate LLM APIs, construct evaluation pipelines, and ship production code — not prototypes.

Claude Code is a daily tool here, not an experiment. You should already be comfortable working with agentic coding tools and using them to move fast without sacrificing quality.

This is a 1099 contract role. Remote, US only. Project-based with flexible scheduling. Compensation is competitive and based on experience.

• Our clients do not need another isolated chatbot prototype. They need AI systems that fit into the operational reality of their business and hold up under production constraints.
• This role helps turn ambiguous client needs into working software: reliable integrations, structured outputs, evaluation loops, and workflows that people can actually trust.
• The right engineer here can move quickly without hand-waving the hard parts: messy source data, model behavior variance, retrieval quality, observability, and human review paths.

• Build and maintain Python services that power client AI integrations
• Integrate with frontier inference APIs across providers: Anthropic, OpenAI, Google (Gemini), OpenRouter, and others
• Design and implement prompt engineering workflows, evaluation pipelines, and RAG architectures
• Ship production features using Claude Code as your primary development environment
• Own scoped deliverables within architectures defined by the engineering lead
• Collaborate directly with the founder and technical lead on active client engagements

• You will work inside existing business systems, not on a blank-slate demo app. That means integrating with live infrastructure, imperfect processes, and real operational constraints.
• Model quality is not assumed. You need to think in terms of evaluation, guardrails, retrieval quality, and failure handling rather than only happy-path prompting.
• Client work requires judgment. You should be able to choose practical solutions, explain tradeoffs clearly, and ship code that can be maintained after delivery.

• Design a retrieval and evaluation workflow for a document-heavy process where hallucinations are unacceptable and source traceability matters.
• Build a multi-provider inference layer that can route around provider-specific limitations while preserving consistent downstream behavior.
• Turn an internal team workflow into a production AI-assisted system with structured outputs, human review steps, and observability for failure cases.

• Python (production-level, not just scripting)
• Hands-on experience integrating frontier LLM APIs (Anthropic, OpenAI, Gemini, or equivalent)
• Prompt engineering beyond basic completion calls: structured outputs, tool use, multi-turn workflows
• Experience building or contributing to evaluation pipelines for LLM outputs
• Working knowledge of RAG architectures: embeddings, vector stores, retrieval strategies, chunking
• Claude Code or equivalent agentic coding tooling as part of your actual workflow
• 1–2+ years of professional software engineering experience
• US-based

• Next.js, React, TypeScript (our client-facing stack)
• Supabase, PostgreSQL (our primary data layer)
• Cloud platforms: AWS, Azure, or Google Cloud
• Data pipeline tooling: dbt, Airflow, Snowflake, BigQuery
• Familiarity with any of our technology partners: Anthropic, OpenAI, Hugging Face, Vercel, Supabase, Snowflake, OpenRouter, Perplexity

• You can talk concretely about systems you shipped, the tradeoffs you made, and what you learned from production behavior.
• You treat agentic coding tools as leverage, not as a substitute for engineering judgment.
• You are comfortable owning a scoped problem, clarifying ambiguity quickly, and delivering something that works in a real environment.

Remote, US only

Async-first with syncs as needed.

1099 contract

Project-based engagement with flexible hours. You manage your own schedule.

Direct collaboration

No layers. You work with the engineering lead, not an account manager.

Production client work

Everything you build ships to real users in real business environments.

• We review each submission ourselves, not through a recruiting layer.
• If there is a fit, the first step is a short technical conversation focused on your past work, judgment, and how you approach production AI engineering.
• No trick questions, no whiteboard theater. We want to understand how you think, how you communicate tradeoffs, and how you actually build.

• Strong applications usually point to real shipped work, not just experiments.
• Be ready to discuss production systems, evaluation decisions, retrieval strategy, and how you use agentic tools in practice.
• A relevant code sample, project, or portfolio link is optional, but it can help us understand your range faster.

Submit your application below. We review every submission. If there's a fit, we'll schedule a short technical conversation — no trick questions, no whiteboard theater. We want to understand how you think and how you work.