BEGIN:VCALENDAR VERSION:2.0 PRODID:IEEE vTools.Events//EN CALSCALE:GREGORIAN BEGIN:VTIMEZONE TZID:US/Eastern BEGIN:DAYLIGHT DTSTART:20260308T030000 TZOFFSETFROM:-0500 TZOFFSETTO:-0400 RRULE:FREQ=YEARLY;BYDAY=2SU;BYMONTH=3 TZNAME:EDT END:DAYLIGHT BEGIN:STANDARD DTSTART:20261101T010000 TZOFFSETFROM:-0400 TZOFFSETTO:-0500 RRULE:FREQ=YEARLY;BYDAY=1SU;BYMONTH=11 TZNAME:EST END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTAMP:20260921T132727Z UID:3BBD132F-F92F-44F2-98F1-3B2DB7D9299E DTSTART;TZID=US/Eastern:20260915T190000 DTEND;TZID=US/Eastern:20260915T200000 DESCRIPTION:Getting the Most Out of Agents Without Trusting Them Blindly — Design Patterns for Agentic AI in Regulated Financial Systems\n\nThis talk is about using agentic AI correctly in regulated systems — places w here a mistake isn't just a bug\, it's a compliance or financial exposure. Three real-world\, before-and-after cases: consolidating custodian data p ipelines feeding trading and portfolio systems\, resolving customer identi ty across the globe with a multi-agent design\, and speeding up loan sanct ioning with a lightweight triage agent. Each one shows where agentic AI ge nuinely earns its cost and where a simpler\, deterministic check already w ins.\n\nThe throughline is a governance-first pattern: agents propose\, bu t a validation layer\, compliance rule\, or human always makes the final c all\, and a model is only invoked when a cheaper check has already failed. Attendees will leave with a practical way to think about using agentic AI in their own regulated or high-stakes systems\, including a tiered\, cost -conscious approach to introducing LLMs\, a self-healing pattern that reso lves repeated pipeline failures automatically\, and a clear-eyed way to de cide when a multi-agent design is worth its added complexity.\n\nLearning objectives\n\n- Apply these patterns to real scenarios to cut manual effor t and prevent token over-usage while getting the most from agentic AI.\n- Decide when a task actually needs an LLM versus a cheaper\, faster check.\ n- Build a pipeline that remembers its own failures\, so repeat errors res olve automatically.\n- Judge when a multi-agent design is worth it — and when one agent\, or none\, is enough.\n\nSpeaker(s): Anubhuti Singh \n\nA genda: \nWEBINAR: 7:00 - 8:00 P.M.\n\nThe Zoom Webinar link and password w ill be forwarded to all registered participants after Noon on the day of t he meeting. Check your spam folder if you don't see the email.\n\nWebinar is open to all.\n\nPDH certificates are available and an evaluation form w ill be emailed to you after the meeting. PDH certificate are sent by IEEE USA 3-4 weeks after the meeting.\n\nVirtual: https://events.vtools.ieee.or g/m/575429 LOCATION:Virtual: https://events.vtools.ieee.org/m/575429 ORGANIZER:ieee@gpamg.org SEQUENCE:41 SUMMARY:Getting the Most Out of Agents Without Trusting Them Blindly — De sign Patterns for Agentic AI in Regulated Financial Systems URL;VALUE=URI:https://events.vtools.ieee.org/m/575429 X-ALT-DESC:Description: <br /><p class="MsoNormal"><strong><span style="mso -bidi-font-weight: bold\;">Getting the Most Out of Agents Without Trusting Them Blindly &mdash\; Design Patterns for Agentic AI in Regulated Financi al Systems</span></strong></p>\n<p class="MsoNormal">This talk is about us ing agentic AI correctly in regulated systems &mdash\; places where a mist ake isn't just a bug\, it's a compliance or financial exposure. Three real -world\, before-and-after cases: consolidating custodian data pipelines fe eding trading and portfolio systems\, resolving customer identity across t he globe with a multi-agent design\, and speeding up loan sanctioning with a lightweight triage agent. Each one shows where agentic AI genuinely ear ns its cost and where a simpler\, deterministic check already wins.</p>\n< p class="MsoNormal">The throughline is a governance-first pattern: agents propose\, but a validation layer\, compliance rule\, or human always makes the final call\, and a model is only invoked when a cheaper check has alr eady failed. Attendees will leave with a practical way to think about usin g agentic AI in their own regulated or high-stakes systems\, including a t iered\, cost-conscious approach to introducing LLMs\, a self-healing patte rn that resolves repeated pipeline failures automatically\, and a clear-ey ed way to decide when a multi-agent design is worth its added complexity.< /p>\n<p class="MsoNormal"><strong>Learning objectives</strong></p>\n<ol st yle="margin-top: 0in\;" start="1" type="1">\n<li class="MsoNormal" style=" mso-list: l0 level1 lfo1\; tab-stops: list .5in\;">Apply these patterns to real scenarios to cut manual effort and prevent token over-usage while ge tting the most from agentic AI.</li>\n<li class="MsoNormal" style="mso-lis t: l0 level1 lfo1\; tab-stops: list .5in\;">Decide when a task actually ne eds an LLM versus a cheaper\, faster check.</li>\n<li class="MsoNormal" st yle="mso-list: l0 level1 lfo1\; tab-stops: list .5in\;">Build a pipeline t hat remembers its own failures\, so repeat errors resolve automatically.</ li>\n<li class="MsoNormal" style="mso-list: l0 level1 lfo1\; tab-stops: li st .5in\;">Judge when a multi-agent design is worth it &mdash\; and when o ne agent\, or none\, is enough.</li>\n</ol>\n<p class="MsoNormal">&nbsp\;< /p><br /><br />Agenda: <br /><p><strong>WEBINAR:</strong>&nbsp\;7:00 - 8:0 0 P.M.&nbsp\;</p>\n<p>The Zoom Webinar link and password will be forwarded to all registered participants after Noon on the day of the meeting. <str ong>Check your spam folder if you don't see the email.&nbsp\;</strong></p> \n<p>Webinar is open to all.</p>\n<p>PDH certificates are available and an evaluation form will be emailed to you after the meeting. PDH certificate are sent by IEEE USA 3-4 weeks after the meeting.</p> END:VEVENT END:VCALENDAR