AI & CLM Consultant · Certified Paralegal

Where Expertise Meets Evidence

I build AI systems for work that can't afford to be wrong, and I prove they hold up before anyone relies on them. Contract intelligence is where it started. The method transfers to any domain where being wrong is expensive.

See the Work Get in Touch
Scroll
100+
Clause detection models built across indemnification, limitation of liability, termination, assignment, and payment obligations
50K
Contracts processed in a single engagement
95%+
Minimum accuracy threshold before any model advances to production
13
Engagements across financial services, energy, and law firm portfolios
How I work

Most legal AI is sold on what it can do. I spend my time on the harder question: proving it did it, on your documents, at a number you can defend.

01 — About

Legal fluency meets AI precision

Chantel Hill
Chantel Hill
AI & CLM Consultant · Certified Paralegal

"Most people on this work are either lawyers who don't trust the model or engineers who've never read a contract. I've done both — and that changes everything about how I build."

Based in
Remote · US
Background
6 yrs legal · 1+ yr contract AI
Credential
Certified Paralegal (CP)
Education
B.S. Psychological Sciences · NAU

Most AI implementations in legal fail for one of two reasons: engineers who don't understand contracts, or lawyers who don't trust AI outputs. I've spent years on both sides of that problem.

With 6 years as a certified paralegal and 1+ year building enterprise data systems, I translate domain risk into model and data requirements — then build, evaluate, and deploy to a 95%+ production standard. Contracts are where I proved it; the method works on any document set that has to be right.

I've worked across financial services, energy, government, and law firm portfolios. I know how source data varies by industry and jurisdiction, how to design pipelines that survive expert scrutiny, and how to present model performance to stakeholders who have never seen a precision-recall curve.

That combination is rare. It's why teams bring me in when the stakes are high, and why the discipline transfers to any data that has to be trusted, not just contracts.

Chantel writes about measurement, judgment, and where this technology should be pointed. The long version →

Clause taxonomy design Extraction pipelines Precision · Recall · F1 CLM implementation Human-in-the-loop design Stakeholder translation
Contract Intelligence at Scale

100+ clause models. 50,000 contracts. 95%+ accuracy.

Detection modeling, extraction pipeline design, and structured dataset creation from unstructured documents — proven on legal language under attorney scrutiny, and built to work on any high-stakes document set.

Agentic Workflow Design

Multi-agent systems with human-in-the-loop governance

Multi-agent orchestration with the patterns enterprise AI deployments need: doer + reviewer agent pairs, structural HITL gates at every handoff, low-confidence escalation, and event traces that show exactly what each agent did. See Stride for a live, working demonstration.

AI Program Delivery & Governance

From use case scoping through production adoption

End-to-end delivery across 13 enterprise engagements: discovery workshops, governance design, pilot programs, change management, and reusable playbooks that compress program setup from days to hours.

Stakeholder Translation

Precision, recall, and F1 in plain language

I present AI performance findings to attorneys, legal ops leaders, and executives — bridging the communication gap most technical teams can't close, and accelerating sign-off where it matters.

02 — Case Studies

Real engagements, real results

Financial Services · Scale

50,000-Contract Portfolio: Building AI Review Infrastructure from Scratch

The Problem

A financial services client needed to extract and analyze key clause provisions across a massive contract portfolio. Manual review at that volume was cost-prohibitive and slow.

What I Built

Designed the full clause taxonomy and extraction architecture. Built and evaluated detection models for indemnification, limitation of liability, termination, assignment, and payment obligations using prompt engineering on Relativity's AI platform.

How I Validated It

Ran precision, recall, and F1 evaluation at every iteration. Built reviewer playbooks and labeling guidelines so attorneys could audit and trust model outputs before any result entered production.

Result: ~50% increase in review capacity on active matters · Models met the 95%+ accuracy bar before attorney handoff · Select clause types reached 100% accuracy
Cross-Industry · 13 Engagements

Building Reusable AI Infrastructure Across Legal Teams

Rather than starting from zero on each engagement, I developed reusable clause libraries, prompt templates, and reviewer playbooks that could be adapted across financial services, energy, and law firm clients — compressing onboarding time and raising the quality floor on every new matter.

13 engagements delivered with consistent production standards · Reusable assets reduced setup time on repeat clause types
Private Sector → Government → AI Consulting

1,500+ Contracts and 200+ Bills a Year: The Foundation That Informs AI Design

Before building AI models, I was the person doing the work they would eventually replace. In private practice: estate planning, probate, M&A, corporate transactions, and real estate closings across 50+ concurrent matters. Then a city attorney's office: 1,500+ contract requests a year at zero backlog on a 24-hour turnaround, plus 200+ bills and resolutions drafted and shepherded to public record. That range — and working across differing court systems and jurisdictions — is the foundation of how I design data systems today.

Breadth most AI consultants don't have: estate planning to M&A to legislation · Fluency in how jurisdictions and court systems differ · Private sector to government office to AI consulting
03 — Built Work  ·  Updated July 2026

Live applications, not just slides

Beyond client engagements, I ship working systems end-to-end — designed, evaluated, and deployed at production standards. Each one is real and working, hosted on chantelhill.com or running live in Google Workspace, and demonstrates a different facet of how I build: multi-agent orchestration, human-in-the-loop governance, workflow automation, and the same evaluation rigor I apply to client clause-detection work.

Multi-Agent System · Live at stride.chantelhill.com

Stride — Multi-Agent Orchestration with Human Oversight

A working AI tool for project managers. Seven specialized agents run in parallel and surface drafts at human-in-the-loop gates before anything advances: status synthesizer, risk detective, meeting prep, action tracker, comms tailor, tracker curator, and a reviewer that critiques every other agent's output. Built on the Anthropic SDK with deterministic curator logic and a status tracker that persists across runs.

What it demonstrates: agentic workflow design, a doer + reviewer pattern that flags low-confidence output before the user sees it, and four kinds of structural human-in-the-loop gates. The same governance discipline enterprise AI deployments require.
See it live →
Causal Eval Harness · Live at attribution.chantelhill.com

Attribution Truth-Checker — Validating AI Claims with Causal Evidence

A working prototype that validates marketing attribution against measured causal incrementality. Built on synthetic data with known ground truth (50 cities, 26 weeks, 100K users), a hand-rolled geo-lift engine using two-way fixed-effects panel regression with cluster-robust standard errors, and a self-evaluation harness scored on precision, recall, F1, and threshold sweeps.

What it demonstrates: the same evaluation rigor I apply to clause-detection work — precision, recall, F1, error analysis, and threshold sweeps — transferred to a different domain to prove the discipline is portable, not industry-specific
See it live →
Workspace Automation · Google Sheets + Apps Script

Project Command Center — Self-Running Project Operations

A six-tab project system that does the manager's prep itself. A live dashboard rolls up KPIs and team load with zero manual entry; a weekly status report writes its own headline and sections from tracker data; a follow-up builder mail-merges ready-to-send stakeholder drafts from any row; and a 7 AM Apps Script digest emails everything overdue or due for follow-up, unprompted. Priority-aware alert logic, 1,400+ live formulas, and every row pre-armed so new work inherits the system automatically.

What it demonstrates: data-ops discipline applied outside the legal niche entirely · automation that hands management time back instead of adding tooling overhead · deliberately built AppSheet-ready, with fields that map cleanly to Jira, Asana, and Monday · status-meeting prep reduced to zero minutes
Request a walkthrough →
04 — How I Think

Decisions I make that most consultants skip

01

I evaluate on precision, recall, and F1 — not just "does it seem right"

Gut-checking outputs isn't evaluation. I run rigorous statistical validation, track error patterns across categories, and document what breaks and why. That's what separates a model that works in a demo from one that holds up on 50,000 real documents.

02

I build human-in-the-loop workflows, not black boxes

AI doesn't replace expert judgment — it has to earn it. Every system I deploy includes confidence thresholds, escalation paths, and audit trail design. Teams need to be able to explain what the system did and why.

03

I scope before I build — and push back when the use case isn't ready

Not every use case is worth building yet. Part of the work is helping teams figure out where automation actually saves time versus where the error rate makes it a liability. Sometimes saying "not yet" is the highest-value thing I can offer.

05 — Writing

Notes on building systems you can actually trust

All Writing
06 — Skills & Stack

The stack behind the work

AI & Delivery
Legal & Ops
Tools
Certifications
AI & Delivery
AI Program Management Cross-Functional Program Coordination Use Case Scoping Client Engagement Leadership Prompt Engineering Clause Detection Modeling Model Evaluation (Precision/Recall/F1) LLM Application Development Agentic Workflow Design Human-in-the-Loop Governance AI Risk & Governance Pilot Program Design Change Management Playbook & SOP Development Workflow & Alert Automation
Tools
Claude Code Cowork Relativity Contracts Pro & GenAI DocuSign CLM & Insight Microsoft Copilot Microsoft 365 Google Workspace Google Apps Script AppSheet Monday.com
Active Certifications
Relativity Generative AI Pro · 2025 Relativity Contracts Pro · 2026 DocuSign CLM Administration Pro · 2026 IBM Prompt Engineering · 2026 Vanderbilt Agentic AI · 2026 Monday.com · 2026 Certified Paralegal (CP)
07 — Work Together

Ready to build contract AI that holds up

If you're scaling AI-assisted contract review and need someone who understands both the legal risk and the technical execution, let's talk.

Send an Email Connect on LinkedIn
chantelhill.cp@gmail.com (623) 243-2883