Mudassir A.

Things I do best: cost-effective agent orchestration that runs reliably in production; making your data pipeline AI-ready; and creating evaluation systems that deliver confidence.

My process:

1. Build from the top → define the expected behaviour and outcome → break it into granular test cases → build and evaluate against those tests.
2. Reduce unnecessary AI decisions wherever deterministic software can do the job, while properly evaluating the parts that need an LLM.
3. Separate retrieval, reasoning, tools, and application logic so individual components can evolve or scale without rebuilding the entire system.
4. Track model and infrastructure costs as part of the architecture, including token usage, model routing, fallback spend, context window utilization, and cheaper alternatives where quality is unaffected.
5. Prioritizing Data Security and Compliance at each increment. (controlling unexpected token spend, which is very common, preventing PII exposure, and maintaining proper credential management)

Please see my portfolio to understand the type of AI applications I can/have built.

I work particularly well on problems involving external data compilation sets and large or complex internal datasets: documents, databases, APIs, SharePoint/Drive content, regulations, operational records, and other internal/acquired business knowledge that needs to become searchable, understandable, or actionable through AI.

𝐑𝐞𝐜𝐞𝐧𝐭 𝐰𝐨𝐫𝐤:

• 𝐀𝐧 𝐀𝐈 𝐜𝐨-𝐟𝐨𝐮𝐧𝐝𝐞𝐫 𝐟𝐨𝐫 𝐬𝐭𝐚𝐫𝐭𝐮𝐩𝐬 𝐚𝐧𝐝 𝐒𝐌𝐁𝐬: specialist agents for planning, fundraising, and go-to-market, plus a graph and vector investor matching engine, exposed via WhatsApp.

• 𝐑𝐞𝐠𝐮𝐥𝐚𝐭𝐨𝐫𝐲 𝐢𝐧𝐭𝐞𝐥𝐥𝐢𝐠𝐞𝐧𝐜𝐞 𝐨𝐯𝐞𝐫 𝐔𝐒 𝐞𝐧𝐞𝐫𝐠𝐲 𝐟𝐢𝐥𝐢𝐧𝐠𝐬: docket timelines, typed extraction contracts, graph database to connect different elements of unstructured documents to map against a user-friendly taxonomical hierarchy and validation gates before anything reaches a count or a report.

• 𝐀𝐳𝐮𝐫𝐞-𝐨𝐧𝐥𝐲 𝐢𝐧𝐭𝐞𝐫𝐧𝐚𝐥 𝐚𝐬𝐬𝐢𝐬𝐭𝐚𝐧𝐭 𝐟𝐨𝐫 𝐚 𝐠𝐥𝐨𝐛𝐚𝐥 𝐛𝐞𝐚𝐮𝐭𝐲 𝐫𝐞𝐭𝐚𝐢𝐥𝐞𝐫: 10,000+ employees, hybrid retrieval over policies and procedures, department-level access control, inside the client's own tenancy under GDPR and CCPA.

• 𝐀𝐧𝐚𝐥𝐲𝐭𝐢𝐜𝐬 𝐚𝐧𝐝 𝐚𝐧𝐨𝐦𝐚𝐥𝐲 𝐝𝐞𝐭𝐞𝐜𝐭𝐢𝐨𝐧 𝐨𝐯𝐞𝐫 𝐢𝐧𝐝𝐮𝐬𝐭𝐫𝐢𝐚𝐥 𝐭𝐞𝐬𝐭𝐢𝐧𝐠 𝐝𝐚𝐭𝐚: text-to-SQL for the records, retrieval for the documentation, and an agent orchestrator deciding which one a question actually needs.