Loading…
RAG Development · Houston, TX
A reservoir engineer at a Houston operator drilling a new Permian well opens the file room (real or digital) to look at analog completions. The relevant material is somewhere in 15,000 well files, a directory of seismic interpretation memos, a folder of frac reports from a contractor that got acquired twice, and a stack of geological narratives written in Microsoft Word in 2008. The question takes a morning.
That is not a headcount problem. It is a retrieval problem. We build RAG systems for Houston energy operators, oilfield services firms, and Texas Medical Center research groups that answer questions over private document libraries in under ten seconds, with citations that point to the paragraph and the source file.
Engagement scope and pricing are set per project, depending on document volume, source count, and the regulatory audit trail your operations group requires.
Tell us which document library your engineers need to query.
A subsurface team at a mid-size Houston independent runs the same lookup forty times a week. Pull the LAS curves for the offset wells, scan the completion reports for stage spacing and proppant loading, cross-reference the geological narrative for the formation top, and compare against the production history. The relevant data sits in four systems with three different metadata conventions and one of them is still partly on a network share. The engineering hours that go into the search dwarf the hours that go into the decision.
On the integrity side, a PHMSA-regulated midstream operator tracks inline-inspection runs, anomaly investigations, engineering critical assessments, and remediation records across thousands of miles of pipe. When the regulator asks about the basis for a recent MAOP recalculation, the answer is sitting in a binder of engineering correspondence and a PDF of the most recent inspection report. The compliance team finds it in an afternoon. A retrieval system finds it in fifteen seconds with a citation.
The Texas Medical Center runs the largest medical complex in the world. MD Anderson alone has a clinical-trial library that spans every major oncology indication, with protocols, informed-consent forms, IRB correspondence, and adverse-event histories going back decades. A principal investigator drafting a new submission spends real time hunting for the analogous prior protocol and the relevant safety data. Houston Methodist and Memorial Hermann run the same retrieval problem on payer contracts and clinical policies for revenue cycle and care coordination.
Generic ChatGPT-style search cannot do any of this. It has no access to your well files, your integrity-management records, or your trial protocols. Industry portals from S&P Capital IQ or IHS Markit answer general questions but do not know your internal completion design, your specific anomaly investigations, or your IRB submission history. A RAG system sits on top of your private corpus and returns the exact paragraph that supports the answer.
Six concrete components, each chosen for the document types common in upstream operations, midstream compliance, and Texas Medical Center research.
LAS curve parser, OCR pipeline for scanned completion reports, python-docx and python-pptx for narrative and seismic-interpretation files. Metadata preserved for API number, field, formation, operator of record, and reporting date so filters run before vector search.
Inline-inspection results, MAOP calculations, anomaly reports, and remediation records ingested with pipeline-ID, segment, and milepost metadata. Designed to answer Part 192 and Part 195 audit questions with cited evidence rather than analyst recollection.
Trial protocols, informed-consent forms, IRB submissions, and adverse-event reports ingested with trial-phase, indication, sponsor, and IRB-number metadata. HIPAA-aligned architecture with role-based access at the chunk level for Texas Medical Center deployments.
BM25 keyword search combined with dense vectors (text-embedding-3-large or domain-tuned alternatives). Metadata filters on basin, formation, well status, or trial phase run before vector search, keeping precision high without sacrificing recall.
Cohere Rerank or BGE Reranker on the top-50 retrieval set raises top-5 precision on technical corpora. Citations render with file name, page, paragraph index, and well or trial ID so a reservoir engineer or compliance officer can paste evidence directly into a memo.
Test set of 100 to 200 real questions from your engineers or research staff with reported recall, MRR, and answer accuracy. Every production query logged with user, chunks retrieved, and answer returned for PHMSA, BSEE, IRB, or HIPAA audit needs.
Houston is the energy capital of the US and a documentation town because of it. ConocoPhillips, Phillips 66, Hess, and the independents run on well files, reserves reports, JOAs, and regulatory correspondence. Halliburton, Baker Hughes, and Schlumberger maintain technical libraries of completion designs, frac-fleet performance data, and field-service histories that customer support and applications engineers query daily. The Port of Houston complex carries customs and hazmat documentation that brokers and operators have to surface on demand. The common thread is that the answer is in a PDF, a LAS file, or a scanned binder somewhere; the bottleneck is finding it.
The regulatory overlay is the second reason a generic search does not work here. PHMSA Part 192 and Part 195 set documented integrity-management obligations on every gas and liquids pipeline operator in the region. BSEE oversees offshore drilling and production in the Gulf and requires Safety and Environmental Management System documentation. EPA Clean Air Act Title V permits drive the petrochemical complexes from Pasadena to Texas City. Every retrieval system we ship into this environment logs the query, the user, and the chunks returned so the audit trail is a byproduct of normal use rather than a separate compliance project.
The Texas Medical Center is the quiet third pillar. MD Anderson, Houston Methodist, Memorial Hermann, Baylor College of Medicine, and the research institutes inside the complex run clinical-trial libraries, payer-contract corpora, and treatment-protocol archives that fit the retrieval shape cleanly. HIPAA, the Common Rule, and 21 CFR Part 11 apply, so we deploy inside your VPC with KMS-managed encryption, role-based access at the chunk level, and full query audit logs from day one.
Tell us what document types you have, roughly how many, and what questions your engineers or research staff need to answer. We reply within one business day with a rough scope and price range.