Fogel Law Group uses IBM watsonx via TrilliCom to accelerate loan-file reviews in the United States; 4-6 hours saved
How Fogel Law Group cuts legal document review from hours to minutes with IBM watsonx
IBM business partner TrilliCom uses IBM watsonx®, IBM Cloud® and IBM data governance technology to help Fogel Law Group organize and review complex loan files in minutes. It also provides lawyers with relevant answers that include citations and adds secure lender portals without deploying separate infrastructure for each client. Commercial real estate and lending transactions produce large, complicated packages of documents. The supporting files can contain scans, tables and records that describe relationships between borrowers, guarantors, property owners, lenders and other legal entities. Before assessing a transaction, an attorney needs to find the relevant language, identify missing documents and verify that the package meets the lender’s checklist. Fogel Law Group, a US law firm focused on commercial lending and real estate transactions, needed to accelerate that work without compromising accuracy or diminishing the client experience. IBM partner TrilliCom addressed the problem with the File Review Agent (FRA), a branded deployment of its TRC AI Platform. FRA ingests and classifies documents, applies configurable checklists, answers questions with citations and tracks each review. Attorneys can now run compliance reviews on complete document sets in minutes instead of hours, dramatically improving efficiency. Retrieval is designed for sub-second latency across thousands of loan documents per transaction. Missing document detection flags gaps before attorneys begin review, while document Q&A compresses initial due diligence by enabling attorneys to ask targeted questions. FRA also uses a multi-tenant architecture: one shared application serves multiple customers while keeping each lender’s data and configuration isolated. New lender portals can be added through tenant configuration rather than deploying new infrastructure. Both the File Review Agent and the underlying deployment platform scale as workloads and clients increase, supporting growth and white-label deployments with complete data isolation. This post examines the review workflow, follows a document through the architecture and shows how IBM watsonx, IBM Cloud and IBM data governance technology divide the work. Before FRA, attorneys reviewed each loan package manually, requiring 12–16 hours of preparation work before substantive legal work begins. They searched across documents for relevant facts and missing files, compared the contents with lender requirements and reconciled names and other entities across records. Automating document acquisition and quality checks can save four to six hours on that manual step alone and improve process performance by 30% to 40%. Attorneys receive complete, validated documents with the required metrics already displayed, compressing much of the pre-review work from hours into minutes while supporting the stricter compliance and governance requirements of SBA lending. With FRA, legal teams configure checklists around lender requirements and loan stages, including whether documents exist, meet quality thresholds or contain matching entities. FRA stores the configuration in a PostgreSQL relational database, enabling teams to update checklists, document classes and review statuses without rebuilding the application. FRA then applies the checklist to indexed loan documents. For each item, it retrieves relevant passages and asks watsonx.ai® to generate a scored answer with source references. The scores can represent the completeness of financial document packages and suggest that further vetting of a borrower is required. Although RAG does not guarantee accuracy, citations let an attorney inspect the evidence and approve or challenge the result rather than accept an unsupported response. For work outside the checklist, attorneys can ask free-form questions about clauses, obligations or relationships through the same document index. TrilliCom designed FRA as a set of connected services for document intake, AI processing, workflow orchestration, storage and security. This modular design enables compute-heavy tasks, including ingest and processing, to run independently of user interactions and other tasks. It also enables multiple operations to be performed in parallel. IBM technologies provide the foundation for processing, retrieving and governing the information throughout the workflow. The architecture diagram organizes the deployment into five domains: users and legal workflows, the FRA application and Legal Research Agent (LRA) services, the document intelligence and agent pipeline, the governed IBM Cloud foundation and verified operational outcomes. The processing path moves from document upload through parsing, embedding and indexing, orchestration, retrieval and answer generation. Separating those responsibilities lets each service scale, fail or change without making the entire review workflow a tightly coupled application. In the first column, Fogel Law Group attorneys and legal staff use a branded, isolated portal for each of their lender clients. In each client’s bespoke environment, Fogel staff upload scanned PDFs, digital files, tables and mixed-format document packages, then start a checklist compliance review, check for missing documents or ask free-form questions. The interface returns scored answers with citations, workflow status and history, while logged actions preserve a traceable record for audit and enable reconstruction of how a conclusion was reached. The application layer controls how attorneys use the File Review Agent and how each request moves through the system. Attorneys work through a graphical user interface, while a secure backend-for-frontend proxy sends authorized requests to a FastAPI backend. The backend manages document ingestion, research, question-answering, background processing, retries and logging. TrilliCom’s Legal Research Agent services connect the user interface, document systems and specialist AI agents. PostgreSQL stores the structured data needed for each review, including lender-specific checklists, document categories, workflow status and audit records. Each lender is treated as a separate tenant, helping keep its data and settings isolated. Documents and vector indexes are stored separately from workflow data, making access controls easier to manage. PostgreSQL also limits request volumes for each tenant to help protect system performance. When documents enter the system (the third column in the diagram), Docling for IBM watsonx prepares them for AI processing. It uses optical character recognition or OCR, to convert scanned pages into machine-readable text. It also identifies page layouts, extracts tables and breaks large files into smaller sections that are easier to search. Watsonx.ai then creates embeddings, which help the system identify passages with similar meaning. It also supports document classification, entity extraction and source-grounded answer generation. Watsonx.data®, together with the Milvus vector database, indexes the document sections and enables semantic search, helping the application find the evidence most relevant to a checklist item or attorney question. IBM watsonx Orchestrate® coordinates specialist agents for tasks such as document classification, checklist review, question-answering and missing-document detection. When an attorney submits a question, retrieval and reasoning agents search the indexed content through retrieval-augmented generation or RAG. The result is returned with citations, allowing the attorney to review the supporting evidence. In simple terms, the process is: parse the document, create embeddings, index the content, route the task, retrieve the evidence and generate a cited answer. Document and AI processing run in the background, so the portal remains responsive while the backend tracks progress, retries recoverable tasks and returns completed results. IBM technologies also provide the storage, runtime and security foundation. IBM Content Manager OnDemand and IBM Cloud Object Storage store incoming files and support auditable retention. IBM Guardium® monitors access to sensitive data, helps enforce security policies and supports audit readiness. IBM Cloud Code Engine runs the containerized services and manages scaling, health checks and software updates. Multifactor authentication, encrypted keys, file validation and access controls provide enhanced protection. Together, these components support practical outcomes for Fogel Law Group. Semantic search and source citations help attorneys move quickly from a question to the relevant evidence. Background processing keeps the portal responsive during complex reviews. Tenant-specific controls help separate lender data, while the shared infrastructure allows the firm to add new lender portals without creating a separate technology environment for each one. As a result, checklist reviews that once took hours can now be completed in minutes. Missing documents can be identified before substantive review begins, while citations and logged AI interactions provide a traceable record of how each result was produced. The architecture supports faster review, greater transparency and a more scalable way to serve lender clients. The largest user-facing benefit is the reduction in time that Fogel Law Group devotes to locating information and checking whether it’s complete. Checklists define what must be evaluated and workflow state shows what remains. RAG supplies relevant evidence for language-model answers. Attorneys can move from a score or summary to the underlying passage, which supports faster review without removing professional judgment from the decision. For both TrilliCom and Fogel Law Group, configuration-led deployment improves total cost of ownership (TCO). Branding, credentials, email settings, workflows and checklists can vary by tenant while the underlying services remain shared. This approach enables Fogel Law Group to onboard another lender without duplicating infrastructure. TrilliCom can apply the same document-processing foundation to other regulated workflows without rewriting the core ingestion, retrieval and governance services. Coupling FRA reviews with watsonx workflows creates a platform for adaptive automation that evolves with changing regulatory and client requirements. With FRA, Fogel Law Group has changed the starting point for commercial loan review. Attorneys can identify gaps, locate relevant provisions and inspect source-backed findings in minutes instead of spending hours searching manually across a file. The configurable platform also gives the firm a repeatable way to support new lender clients without building a new application for each one. The real value isn’t just receiving an AI-generated answer. It’s being able to reach the relevant evidence faster, understand what’s missing and dedicate time to the legal issues that require professional judgment. TrilliCom selected IBM technology to unite document processing, enterprise AI, data retrieval, orchestration, security monitoring and governance in one foundation. Built-in governance was critical for SBA lending practices such as Fogel Law Group’s, where strict compliance requirements apply. Loan records must remain available and unaltered throughout terms that can span 15–30 years. IBM content management and cloud storage services support that long-term record integrity, while the broader architecture enables faster legal reviews with fewer errors. TrilliCom is now exploring how the same foundation is positioned to support expanded stages of lender due diligence and other document-intensive legal workflows.