Founding Engineer, PhrasIQ

Location: Remote (or hybrid, open to discussion)
Stage: Pre-seed, equity-only until seed close or ARR milestone

What PhrasIQ is

PhrasIQ is an agentic FP&A and decision intelligence platform for enterprise finance teams. The core bet is that finance teams won't trust AI-generated numbers unless every output is auditable, traceable, and explainable back to source data. We call this reducing "alignment latency": the gap between an AI producing an answer and a human being willing to act on it.

The product is built on two layers:

1. Loktak — the deterministic backbone. Versioned finance ontology, canonical finance objects, data lineage, eval scoring. This is the system of record and the system of truth.
2. InsightOS — the orchestration layer. Agents, forecasting, narrative generation, UI. This is where the AI lives.

One rule governs everything: InsightOS never touches Loktak's database directly. It only ever goes through Loktak's API. No shortcuts, no unified credential set, no "just this once" direct query. If that constraint doesn't make immediate sense to you, or sounds like premature architecture for a pre-seed company, this probably isn't the right role.

We're hiring two founding engineer tracks. You can apply to either, or tell us which one you'd naturally gravitate toward and we'll figure it out together.

Track 1: Founding Platform Engineer (Loktak)

You'll build the deterministic backbone that everything else depends on.

What you'll own:

• Versioned finance ontology and canonical object models
• Data lineage: every number traceable to its source, every transformation logged
• Eval scoring infrastructure for AI-generated outputs
• The API surface that InsightOS is permitted to call — designing it to be strict enough that it's structurally impossible to bypass

You're a strong fit if:

• You've built systems where correctness and traceability mattered more than speed of iteration (fintech, healthtech, compliance-adjacent, data infrastructure)
• You think about schema design and versioning before you think about UI
• When someone describes a system with a "trusted" internal shortcut, your instinct is discomfort, not relief

Track 2: Founding AI/Agent Engineer (InsightOS)

You'll build the orchestration layer: the agents, the forecasting logic, the narrative generation, all of it operating strictly within the boundaries Loktak defines.

What you'll own:

• Multi-agent architecture across four scoped agents (Grounding/Retrieval, Modeling/Forecast, Variance/Anomaly, Narrative/Drafting), each with its own least-privilege credential scope into Loktak
• Forecasting and decision-brief generation logic
• The user-facing product surface

You're a strong fit if:

• You've built with LLM agents/orchestration in production, not just prototypes
• You default to asking "what should this agent NOT be able to touch" before "what should this agent be able to do"
• You've worked somewhere that made you care about explainability, not just output quality

What's true for both tracks

Comp: Equity-only until we close our seed round or hit ARR milestone, whichever comes first. We're telling you this now, not five interviews in, because we'd rather you self-select in or out early than waste your time or ours.

Stage reality: You will be the 2nd or 3rd engineer at the company. There's no scaffolding. You'll be making architectural calls that outlast this hiring round.

Path: For the right person on either track, there's a real path to a co-founder-level role, not just an early employee slot.

What we're actually testing for: Less "do you know the right frameworks," more "when you hit ambiguity, do you treat it as a product problem or just an engineering problem." The strongest people we've talked to instinctively ask who the output is for before they ask how to build it.

Process

1. Intro call (30 min) — mutual fit, equity structure surfaced up front
2. Technical deep dive (60-90 min) — architecture walkthrough, including a boundary/traceability scenario specific to your track
3. Take-home — mirrors an actual PhrasIQ problem (e.g., a Decision Brief Generator), intentionally underspecified in places
4. Founder conversation — vision, working style, path forward

Before you apply here, please send us answers to the following prompt under "What interests you about working for this company?":

• What is a project you built that had to be correct, rather than just fast?

• What is one system you have worked on with a boundary or permission model you are proud of (or one you wish had existed)?

• Which track pulls at you more, and why?