Back to Case Studies
Oil & Gas Document Processing AI-Powered OCR

Fuel Delivery Log Digitization

How Colville accelerated their digital transformation with AI-powered document processing on Alaska's North Slope—delivered 35 days after Gemini 3 Pro launched.

35

Days to Production

10-24

Pages Per Day

-40°F

Field Conditions

EPA

Compliance Ready

About Colville, Inc.

Colville is a fuel services company operating on Alaska's North Slope, providing critical fueling operations for oil and gas production in one of the most demanding environments on Earth. Their team delivers fuel to equipment and facilities across remote Arctic locations, supporting major energy producers with reliable service in extreme conditions.

The Challenge

Modernizing Field Data Capture

Colville's fuel delivery operations generate 10-24 pages of handwritten logs daily. Field workers record fuel dispensed to equipment across remote Arctic locations, often in extreme conditions—temperatures reaching -40°F, heavy gloves, limited visibility. These logs must be digitized for EPA air monitoring compliance requirements.

As part of their broader modernization strategy, Colville identified an opportunity to streamline this manual process. The traditional workflow required pulling staff from other roles for manual data entry—time that could be better spent on higher-value work.

Unique Considerations
Multiple form variants from different service routes
Varied handwriting styles across field personnel
Circled location codes requiring contextual interpretation
Documents scanned at various angles and orientations
Field conditions producing time-pressured entries

The Approach

Colville had already experimented with AI tools enough to know that modern models could probably handle the messy handwriting and inconsistent formats. But bridging the gap between "this should work" and "this actually works in production" required more than feeding documents into ChatGPT.

In mid-November 2025, Delve Group was tracking Google's upcoming Gemini 3 Pro model, which showed promising OCR capabilities in early benchmarks. After discussing the fuel log challenge with Colville, we agreed to run a proof-of-concept as soon as the model launched.

The following week, Gemini 3 Pro was released. Using a sample of real-world documents, we validated that the new model could handle even the most challenging handwriting and form variations. The results were strong enough to move forward immediately.

35 days after Google released Gemini 3, Colville had a production-ready tool in their hands.

This timeline illustrates Delve Group's approach: we work in the gap between what AI models are capable of and what off-the-shelf tools allow you to do. The technology existed, but turning it into a reliable, production-ready solution required expertise in prompt engineering, system design, and understanding the specific operational context.

The Solution

Delve Group developed a web-based OCR tool leveraging Google's Gemini 3 Pro AI model to intelligently transcribe handwritten fuel delivery logs. The solution was designed for rapid deployment while maintaining accuracy through smart validation workflows.

Intelligent Transcription

Drag-and-drop PDF upload with AI-powered recognition of handwritten entries, including challenging handwriting. Context-aware interpretation using location lists and contractor data with automatic form variant detection.

Human-in-the-Loop Validation

Flagging system for uncertain transcriptions with side-by-side comparison. Easy correction interface for edge cases and progress tracking across multi-page documents.

Data Management

Configurable alias system for contractor and location names. CSV export for integration with existing systems. Secure, cloud-hosted platform accessible from any browser. Cost tracking for Gemini API usage.

Technical Approach

Rather than building a complex training pipeline or traditional OCR engine, Delve Group leveraged Gemini 3 Pro's multimodal capabilities with carefully crafted instructions:

1

Contextual Grounding

Send documents directly to Gemini with known locations, contractors, and equipment types embedded in the prompt

2

Structured Output with Exception Flags

Return transcribed rows of structured data with flags noting illegible entries, unusual values, or missing fields

3

Human Review

Surface flagged items for quick validation and correction

The Results

Immediate Impact
High-Volume Processing

Handles 10-24 pages of handwritten logs daily with ease.

Sustainable Daily Workflow

What previously required dedicated staff time now fits into the normal workday.

Accuracy Maintained

Human validation ensures data quality while AI handles the heavy lifting.

Rapid Deployment

Solution delivered and operational within 35 days of model release.

Operational Benefits
Staff can focus on core responsibilities instead of tedious data entry
Near real-time data availability for compliance reporting
Reduced risk of transcription errors from manual entry fatigue
Foundation established for continued digital transformation
"Delve Group understood exactly what we needed: something simple that just works. They delivered in three weeks what could have taken months and cost three times as much. The tool handles even our most challenging documents, and now we can demonstrate real value to our clients. Having a local team that understands our industry made all the difference."

Mark Warrell

Colville

Why Delve Group

Practical AI

Applying cutting-edge technology to solve real operational problems.

Rapid Deployment

Delivering working solutions in weeks, not months.

Human-Centered Design

Keeping people in the loop where judgment matters.

Alaska Expertise

Understanding the unique challenges of operating in remote, harsh environments.

Foundation for Growth

Building solutions that can expand and evolve with client needs.

Project Details

Timeline

Nov - Dec 2025

Industry

Oil & Gas / Fuel Services

Location

North Slope, Alaska

Tech Stack

Gemini 3 Pro, Azure

Ready to digitize your operations?

Let's discuss how AI can streamline your document processing workflows.

An unhandled error has occurred. Reload 🗙