AI Engineer · Agentic Giants

Ahmad Ishfaq

Ahmad Ishfaq is an AI Engineer at Agentic Giants, an AI agent development company in Aubrey, Texas. He implements the retrieval, evaluation, and MCP-tool layers behind the production AI agents Agentic Giants ships. Ahmad also reviews the measurement protocol published on /proof before it goes live.

What Ahmad Ships on Every Engagement

  • Production GraphRAG pipelines on Neo4j. Ingest, entity linking, retrieval, and provenance layers for the enterprise knowledge graphs behind the 90% unsupported-answer reduction figure on /proof.
  • Evaluation harnesses that survive after we exit. Every engagement ships with a golden test set and a regression harness the client’s team can run after Agentic Giants leaves the code review rotation.
  • MCP server implementation and audit logging. The tool layer that gives AI agents governed access to enterprise systems, with least-privilege scoping and immutable per-action traces. See MCP Server Development for the delivery pattern.

Areas of Expertise

GraphRAG pipeline engineering on Neo4j

Ahmad Ishfaq builds production GraphRAG pipelines on Neo4j — the retrieval and provenance layers that make an AI agent's answers citable back to specific entities in the knowledge graph. Ingest, entity resolution, hybrid retrieval, and evaluation instrumentation are all in scope; the goal is one shipped pipeline with a running regression harness the client's team can maintain.

See /services/graphrag-implementation for the delivery pattern and /proof for the measured impact on unsupported-answer rate.

AI evaluation harnesses for production systems

Ahmad Ishfaq designs and implements evaluation harnesses that score AI agent outputs against a golden test set before a release ships. The harness runs on every commit, blocks deploys on regressions, and produces the numbers the /proof page cites. A production AI system without an evaluation harness is a demo; the harness is what makes reliability engineered rather than hoped for.

This is the discipline behind the 500-query enterprise evaluation set on /proof#unsupported-answers.

MCP server implementation and immutable audit logging

Ahmad Ishfaq implements Model Context Protocol servers with least-privilege tool scoping, human-in-the-loop approval on state-changing calls, and immutable per-action audit logs. The MCP layer is what allows an AI agent to move money, update records, or file claims inside a regulated environment without the policy layer becoming something the model can talk its way past.

See /services/mcp-server-development for the standard implementation pattern.

Where Ahmad’s work appears on the site