Work — Arash Askari | Agentic AI & Enterprise Systems

Context: One agentic reference architecture supports four finance POCs for contract capture, invoice chat, document links, and billing checks.
Decision: Use one agentic reference architecture for four finance workflows. Do not build one general assistant.
Work: The reference architecture covers contract capture, billing, and invoice chat.
The invoice agent answers questions about a selected invoice.
The upload interface captures fields from MSAs, SOWs, POs, and invoices.
The billing agent checks each invoice against its project documents.
Outcome: Four production POCs covered contract capture, invoice chat, document links, and billing checks.
The shared design gives future finance agents the same controls and evidence model.
Stack: Azure · Reflex · LangChain · LLM agents · MCP
Constraints: Anonymized pharmaceutical environment · Financial documents · Evidence that a person can verify

Context: The migration plan covers more than 15,000 SQL objects, their dependencies, compatibility, and performance impact.
Decision: Map every Synapse dependency before you define the Fabric target or migration order.
Work: The current and target designs show each data pipeline change.
The assessment covers more than 15,000 SQL objects in Azure Synapse.
The migration report records dependencies, compatibility, and performance impact.
Outcome: The migration plan covers dependencies, compatibility, and performance impact across more than 15,000 SQL objects.
Stack: Azure Synapse · Microsoft Fabric · Python
Constraints: Large existing estate · Migration continuity · 15,000+ SQL objects

Context: One qualified CI/CD and release process controls more than 70 Snowflake accounts.
Decision: Treat release engineering and account governance as one qualified Snowflake operating model.
Work: The CI/CD and release process meets GxP controls.
One governance workflow covers more than 70 Snowflake accounts.
Account owners can see each change and its rollback path.
Outcome: One CI/CD and release process now serves more than 70 accounts.
Owners can see each change, rollback path, and account assignment.
Stack: Python · Azure · Snowflake · AWS
Constraints: GxP-regulated environment · 70+ accounts · Auditability and rollback

Context: One AWS platform controls Tableau Bridge networking, access, and deployment.
Decision: Run Tableau Bridge as one governed AWS platform. Replace manually operated connectors with managed network, access, and deployment controls.
Work: Teams use one self-service Tableau Bridge platform.
The platform controls network access and automated deployment.
Outcome: Teams can use the same cloud platform without managing separate connectors.
Stack: AWS · Tableau
Constraints: Global pharmaceutical environment · Network boundaries · Access control

Context: The work improved a production Snowflake warehouse and added CI/CD for graph database integration.
Decision: Improve the production warehouse while you establish a controlled CI/CD path for graph integration.
Work: The work made functional and operational improvements to a production Snowflake warehouse.
The team built a CI/CD pipeline for graph database integration with the data and analytics platform.
I advised on data warehouse and analytics vendor selection.
Outcome: The work produced functional and operational improvements to the Snowflake warehouse.
The work produced a CI/CD path for graph integration and evidence for vendor selection.
Stack: AWS · Snowflake · CI/CD
Constraints: Production data platform · Integration continuity · Vendor selection

A hard system problem?

Start with the decision. Then prove it.

Send me the decision, constraint, or failing system. I’ll tell you where I would start.