Loading…
RAG Development · Dallas, TX
A senior landman at a Permian-focused operator spends most of Tuesday in a SharePoint folder structure that nobody has maintained since 2014, looking for a 1998 surface-use agreement that controls whether a new pad can spud. A compliance analyst at a Texas LDC pulls up four PHMSA dig sheets from three different vendors to answer a single question about a single anomaly. A carrier-management attorney at AT&T greps a wholesale agreement library that lives in five document management systems.
The corpus is not the problem. The retrieval over the corpus is the problem. We build RAG systems for Dallas-area energy, telecom, insurance, and healthcare firms that answer questions over your own document library in under ten seconds, with citations that point to the specific page and paragraph.
Engagement pricing is scoped to fit, depending on document volume, source systems, and the audit and regulatory requirements your team operates under.
Tell us which document library is eating your team's week.
The upstream majors with a meaningful Texas presence (ExxonMobil in Irving, ConocoPhillips, Pioneer, the producing arm of Hess inside Chevron, the various independents that survived the 2020 shakeout) inherited document libraries through forty years of acquisitions. Well logs from 1986. Lease agreements with heirs you can no longer find. Seismic reports referencing surveys run by companies that no longer exist. The Texas Railroad Commission filings tied to each of these sits in another system. When a regulatory request lands or a divestiture goes into diligence, the question is never whether the document exists, it is whether anyone can find it inside the available window.
AT&T's carrier-contracts corpus tells a parallel story. Tens of thousands of interconnect agreements, wholesale MSAs, rate exhibits, and PUC filings, layered through twenty years of mergers (SBC, BellSouth, the Cingular wireless consolidation). When a regulatory team needs to pull every contract with a specific MFN clause or a specific access charge methodology, the existing answer is a paralegal team and a deadline.
Healthcare systems run into the same wall on a different axis. Tenet Healthcare and Baylor Scott & White each maintain clinical policy libraries spanning thousands of documents, annual payer contracts with United, Aetna, Blue Cross, and Cigna, and Texas Medicaid program rules that change quarterly. A revenue-cycle director asking whether the 2025 Aetna contract permits a specific out-of-network billing path is doing the same retrieval task as a landman looking up a pugh clause, with the added wrinkle that the answer has to be auditable for Texas Department of Insurance review.
Generic ChatGPT cannot answer any of these questions. It has no access to your documents. Vendor platforms like Quorum (for energy land), Salesforce CLM (for contracts), or Epic (for clinical) answer their own questions but do not bridge corpora. A RAG system sits on top of your private library and returns the exact paragraph that supports the answer, with the citation a compliance officer or auditor can paste into a memo.
Six components chosen for the document types and regulatory environment of Texas energy, telecom, insurance, and healthcare.
Native PDF text where the source is clean, AWS Textract or unstructured.io for scanned filings, and a layout-aware vision model for older well logs and dig sheets. Each document carries domain metadata (API number, milepost, counterparty, signed date) so filters run before vector search.
Ingest ILI runs, dig sheets, annual inspection reports, and HCA assessments with metadata for asset ID, segment, anomaly type, and class location. Query logs retained for the PHMSA documentation window so the system itself is auditable rather than another shadow source.
Carrier interconnect agreements, wholesale MSAs, and energy lease agreements chunked at the clause level with parent-section metadata. A query about a most-favored-nation provision or a pugh clause returns the exact paragraph with article, section, and definition cross-references intact.
BM25 plus dense vectors (text-embedding-3-large by default, Voyage AI for finance and legal corpora). Metadata pre-filters on county, formation, operator, RRC filing date, or insurance code section keep precision high without sacrificing recall on multi-system queries.
Every retrieved chunk renders with source document name, filing or signed date, page, and paragraph index. The format pastes directly into a Texas Department of Insurance audit response, a PHMSA inquiry reply, or an internal credit memo without manual reformatting.
Deploys inside your VPC or Texas-region tenant with Active Directory or Okta SSO. Chunk-level access controls respect existing document permissions. Every query is logged with user, timestamp, chunks retrieved, and answer returned to satisfy internal audit and external regulatory review.
Dallas-Fort Worth is a regulated-industry town. ExxonMobil runs its global business from Irving. AT&T headquarters and its regulatory function sit downtown. Texas Instruments, Jacobs Engineering, McKesson, and the corporate side of Frito-Lay in Plano each carry the document apparatus that goes with their scale. Southwest Airlines runs from Love Field with an FAA-driven technical-publications library. Match Group operates trust and safety policies across half a dozen consumer apps. The common thread is that the answer to any operational question is sitting in a PDF or a SharePoint file somewhere, and the bottleneck is finding it.
The regulatory overlay is heavier here than in most metros. The Texas Railroad Commission governs every well and pipeline. PHMSA imposes federal rules on top. The Texas Department of Insurance regulates the property and casualty market that USAA (in nearby San Antonio), Allstate's Dallas operations, and the various captive insurers actually run. Texas Medicaid and the Texas Health and Human Services Commission set rules that Tenet and Baylor Scott & White have to operationalize across hundreds of facilities. None of these regulators care about your DMS architecture; they care about whether you can answer their question with a citation.
Insurance is the quiet third vertical. Dallas hosts the captive and reinsurance functions for several Fortune 500 operators, and the property and casualty side of the market here runs on policy-form libraries, regulatory bulletins, and claim-precedent files. A claims supervisor asking which Texas hailstorm losses from 2023 to 2024 referenced a specific exclusion endorsement should not need to open seven systems to answer. Retrieval over the policy and claim corpus is exactly the shape of work where the ROI is measurable in the first month.
Tell us what document types you have, roughly how many, and what questions your team needs to answer. We reply within one business day with a rough scope and price range.