Applied AI and Agent Systems Engineer (Part-Time)

Build the reasoning systems that make Tegy materially more useful than a generic AI chat product. You will work on agent behavior, retrieval, structured analysis, evaluation, reliability, and the product loops that turn complex business context into work a senior strategy team could defend.

This is applied product engineering. The goal is not a clever demo or a parade of model calls. The goal is a system that produces useful decisions and artifacts consistently enough for real business work.

About Tegy
Tegy is building the strategy team ambitious companies wish they had. It turns messy business context into defensible decisions, models, memos, decks, and action plans with senior-grade rigor on demand.

The technical challenge sits at the intersection of domain reasoning, human judgment, product design, and modern AI systems. The people who join now will help define the methods, evaluation standards, and reliability bar behind the product.

Why join now
• Apply AI to consequential work where reasoning quality can be tested, not merely showcased.
• Shape the agent architecture and evaluation system before conventions harden.
• Work directly with founders, strategists, product leaders, and early users.
• See your work move quickly from experiment to customer workflow.
• Build the technical and product moat behind an AI-native strategy platform.

What you will own

1. Reasoning and agent workflows. Design systems that decompose ambiguous business questions, choose useful tools, maintain context, and produce structured outputs.
2. Retrieval and context quality. Improve how customer information, source material, prior work, and domain knowledge are selected, grounded, and traced through an analysis.
3. Evaluation. Build representative datasets, rubrics, automated checks, and human review loops that measure usefulness, rigor, consistency, and failure modes.
4. Reliability and observability. Diagnose model, prompt, tool-use, latency, cost, and orchestration failures and make the system easier to trust in production.
5. Product integration. Partner with full-stack engineering and product to turn AI capabilities into clear user workflows rather than isolated backend experiments.

What success looks like
First 30 days: Understand the current reasoning flows, reproduce the most important failure modes, establish an evaluation baseline, and ship one measurable improvement.

By 60 days: Own an agent or retrieval workflow in production with instrumentation, quality thresholds, and a clear feedback loop from users and reviewers.

By 90 days: Improve a decision-relevant quality metric, reduce a meaningful reliability or cost failure mode, and establish an evaluation pattern the team can reuse across new capabilities.

Who will thrive
• You have built and operated LLM, retrieval, or agent systems beyond prototypes.
• You can move between experimentation and production engineering without treating either as somebody else's problem.
• You design evaluations before declaring an improvement and can explain what a metric misses.
• You understand structured outputs, tool use, observability, guardrails, latency, cost, and human review as parts of one system.
• You are curious about strategy and business decision-making and want to translate domain methods into product behavior.

Strong candidates may come from applied AI, machine learning engineering, AI product engineering, information retrieval, developer tools, or domain-heavy workflow automation. Published work is welcome, but production evidence matters more than credentials alone.

Engagement and compensation
Start with a paid four-to-six-week technical sprint or a fractional contract of roughly 10 to 20 hours per week. The sprint range is US$3,000 to US$8,000. Ongoing work typically falls in the US$35 to US$70 per hour range, depending on scope, availability, location, and demonstrated capability.

The equity range is 0.0% to 3.0%, with the specific grant shaped by ownership, sustained contribution, commitment, and demonstrated impact.

This role is open worldwide. You should be able to maintain reliable scheduled overlap with US Eastern time for technical decisions and collaboration.

How to apply
No long cover letter is needed. Send:
• One or two AI or agent systems you personally built, including the user outcome, your contribution, and the hardest failure mode you addressed.
• A short response: if an AI strategy product produces fluent but inconsistent recommendations, what would you measure first and how would you improve it?
• Your location or time zone, availability, preferred engagement model, and compensation expectations.

Candidates who advance will discuss real system tradeoffs; any substantial build exercise will be paid.

Exceptional people rarely arrive on a hiring calendar. We meet strong candidates year-round and move quickly when there is a compelling fit.