The Client
Collide, built by Digital Wildcatters, is a knowledge platform for the oil and gas industry. Its enterprise customers are energy companies sitting on decades of documents.
The Problem
Those customers had 100,000+ documents: drilling reports, AFEs, well logs, spreadsheets, slide decks. Finding one number meant an analyst digging through PDFs for hours, and decisions waited on the digging. Existing search tools were slow, missed oil and gas terminology, and nobody trusted what they returned.
What I Built
I built the AI search platform end to end: the pipeline that reads and indexes every document, the chat interface people use to ask questions, and the dashboards that show the team what’s happening. You ask in plain English and get the answer in seconds, with a link to the exact page it came from. If the answer isn’t in the documents, it says so.
The system speaks oil and gas, so a question about AFE overruns finds the right documents even when they never use those exact words.
Under the hood
End-to-end Azure ML ingestion pipeline in Python using Document Intelligence, Computer Vision, and Cognitive Search. Fine-tuned PyTorch transformer embeddings for oil and gas terminology, with citation-linked retrieval grounding every answer in source documents. Apache Kafka for distributed ingestion under upload spikes. Ruby on Rails and React frontend with native websocket chat and analytics dashboards. Docker and Kubernetes orchestration. FastAPI and Prometheus telemetry tracking ingestion throughput, request latency, and pipeline failure rates. 90% test coverage across pytest, Jest, and RSpec, deployed through Jenkins and GitHub Actions.
The Result
- 3x faster document retrieval across the full 100,000+ document library
- 5 enterprise energy clients running on the platform at delivery
- 90% test coverage, so the fragility of the legacy system never came back
- Platform stability and adoption contributed to Collide securing $5M in seed funding