AI agents and workflow automation

A customer investigation can require reading messages, checking several systems, applying policy and taking an action. A document review needs the right sources, an assessment against business requirements and a result someone can use. Automation has to connect those steps to the process already in place.

I work with the people doing and managing the work to identify what should change, choose the appropriate combination of software and models, and implement it. My delivery experience includes production contract intelligence and live trading platforms, alongside the demonstrations and software described here.

I start with your procedures, existing outputs and actual cases, including exceptions and work that gets reopened, and walk through them with the operators, managers and people receiving the results. Their decisions reveal requirements that a process diagram or a policy document can leave out.

I translate that understanding into the system design: the information it needs, the decisions a model can make, the rules code enforces and the points where a person decides. I develop the integration and application code with your engineers, or take responsibility for a defined implementation, including testing, recovery and operating ownership.

Document systems can extract fields and tables for downstream processing, answer questions over a collection, or review a complete packet against business requirements. The design follows the task. A searchable passage can answer a question; reviewing an agreement requires knowing which versions and attachments belong to it and which requirements must be examined.

My delivery experience includes a production contract-intelligence system connecting source documents to structured interpretation and business rules. In a new implementation, the same practical questions arise: which sources govern, how findings link to evidence and how the person reviewing a result can correct it.

Parser selection is one part of this work. PDF-Bench provides a task-specific comparison of parsing results, while PDFsmith supplies document-processing interfaces. They support the information-processing step within the larger document workflow.

A procurement analyst needs an NDA assessed against company requirements. The Contract Review demonstration shows an agent assembling an exception memo from the agreement, competing amendment copies and an internal policy. The memo identifies a backup-retention exception and a missing security exhibit, with references to the documents behind each finding.

Changing the proposed use of customer data changes the review. Following that variation shows how a requirement affects the agent's investigation and the decisions it sends to Legal or Security. The demonstration makes the path from documents to a usable review concrete.

The Customer Support demonstration investigates a replacement that never arrived after the original return window closed. The agent reads order and shipment records, applies the remedy policy, requests an exception decision and confirms the customer's choice. It then completes the refund in the demonstration environment.

The model chooses which records to examine and what to ask next. Application code enforces ownership, approval and the amount that can be refunded. Variations show a customer declining the refund and a response being lost after the refund was recorded. They explain why an agent needs both flexible investigation and reliable control of system actions.

I build the evaluation around the work the system must complete. For document review, that means coverage and whether the findings follow from the sources. For a system that takes actions, I check the resulting records, policy requirements and recovery from interrupted work, and include unresolved cases and the effort people spend reviewing and correcting results.

I develop representative cases, acceptance criteria and evaluation tools alongside the application. Review arrangements follow the consequences of error and measured performance: some decisions need prior approval, while others can use exception handling and sample-based checks. Existing systems can be assessed and improved using the same approach.

I compare ordinary software, a fixed workflow with model-assisted steps and an agent on their complete operating plans: implementation, processing, review, correction, exceptions and maintenance, on the same workload and quality basis.

That comparison identifies where further development would help. Better source data, a direct integration or a clearer handoff may improve the process more than another model call. The measure is whether the change releases capacity, reduces spending or improves the service, including the work needed downstream.

My financial-markets work includes quantitative research and live trading platforms used in production over multiple years, with execution, risk controls and operating responsibilities. That experience informs how I approach integrations, persistent state, monitoring and recovery in business applications.

In Machine Learning for Trading, the implementation treatment extends to research and forecasting agents, retrieval over financial documents, and evaluation. The accompanying Agent Lab presents dated forecasts and reasoning from an operating forecasting system.

AI-assisted development also needs a defined task, a working test environment and review of the resulting change. The Coding Agent Toolkit supports that workflow through specification, implementation, verification and continuation across sessions.

The coding-agent article develops a worked change from an initial defect through tests and a corrected patch. It complements the business-agent demonstrations by examining how the software itself is developed and checked.