https://irisseaman.com/how-i-work-with-others/2026-10-10T22:53:21+00:00weekly0.6https://irisseaman.com/2026/10/09/production-ai-assistant-employee-knowledge-operations/https://irisseaman.com/wp-content/uploads/2026/10/employee-ai-assistant-architecture-evolution.pngEmployee AI assistant architecture evolutionThe architecture grows only when a new requirement establishes the need.https://irisseaman.com/wp-content/uploads/2026/10/production-ai-engineering-responsibility-map.pngProduction AI engineering responsibility mapModel behavior and system enforcement meet through explicit, testable contracts.https://irisseaman.com/wp-content/uploads/2026/10/employee-assistant-end-to-end-pipeline.pngEmployee AI assistant end-to-end pipelineThe online request path and the offline evaluation, planning, and improvement loop.https://irisseaman.com/wp-content/uploads/2026/10/employee_assistant_production_architecture.pngEmployee AI assistant production architectureThe employee sees one conversational interface. Behind it, the architecture separates request identity, reasoning, knowledge, tools, execution, and operations.2026-10-10T22:24:49+00:00monthlyhttps://irisseaman.com/current-research/https://irisseaman.com/wp-content/uploads/2026/09/img_4665-copy.jpgIMG_4665 copyhttps://irisseaman.com/wp-content/uploads/2026/09/crystal_edge_schematic.pngcrystal_edge_schematichttps://irisseaman.com/wp-content/uploads/2026/09/graph_on_mri_radiologist_view.pnggraph_on_mri_radiologist_viewhttps://irisseaman.com/wp-content/uploads/2026/09/ispy2_two_patient_rollout_static.pngispy2_two_patient_rollout_statichttps://irisseaman.com/wp-content/uploads/2026/09/endpoint_calibrated_spatial_forecasting.pngendpoint_calibrated_spatial_forecastinghttps://irisseaman.com/wp-content/uploads/2026/09/endpoint_calibrated_spatial_forecasting.jpgendpoint_calibrated_spatial_forecastinghttps://irisseaman.com/wp-content/uploads/2026/09/registered_10_color_composite-edited.pngregistered_10_color_compositehttps://irisseaman.com/wp-content/uploads/2026/09/registered_10_color_composite.pngregistered_10_color_compositehttps://irisseaman.com/wp-content/uploads/2026/09/graph_on_mri_radiologist_view-4.jpggraph_on_mri_radiologist_viewhttps://irisseaman.com/wp-content/uploads/2026/09/ssl_embedding_model_design.jpgssl_embedding_model_design2026-10-10T22:24:28+00:00weekly0.6https://irisseaman.com/production-ai/2026-10-10T22:05:11+00:00weekly0.6https://irisseaman.com/2026/10/10/crystal-shell-graphs-breast-mri-forecasting-uncertainty/https://irisseaman.com/wp-content/uploads/2026/10/crystal-shell-tumor-graph-forecasting-v2.pngCrystal-Shell Tumor Graph ForecastingFrom pretreatment breast DCE-MRI to a regional presurgery forecast and patient-conditioned uncertainty.https://irisseaman.com/wp-content/uploads/2026/10/crystal-shell-tumor-graph-forecasting-breast-mri.pngCrystal-Shell Tumor Graph ForecastingFrom pretreatment breast DCE-MRI to a regional presurgery forecast and patient-conditioned uncertainty.https://irisseaman.com/wp-content/uploads/2026/10/manuscript-figure-5.pngUncertainty interval tradeoffs (manuscript Figure 5)Manuscript Figure 5. The train-selected calibration improved held-out interval score and coverage while increasing mean interval width; the held-out CQR ranking was not a train-selected result.https://irisseaman.com/wp-content/uploads/2026/10/manuscript-figure-3.pngGraph-model controlled comparison (manuscript Figure 3)Manuscript Figure 3. Source-held-out model-variant comparison on the nested ACRIN-6698 subset. The chart uses Monte Carlo mean MAE, not deterministic MAE.https://irisseaman.com/wp-content/uploads/2026/10/manuscript-figure-1.pngCrystal-shell graph forecasting pipeline (manuscript Figure 1)Manuscript Figure 1. Research pipeline. The separate pCR classifier also uses observed follow-up features; it is not a pretreatment-only classifier.2026-10-10T22:04:52+00:00monthlyhttps://irisseaman.com/2026/10/07/agentic-ai-researcher/https://irisseaman.com/wp-content/uploads/2026/10/twin_forge_current_architecture_v2.pngThe current Twin Forge research architecture, revisedThe agents exchange specifications and evidence through the repository. Obsidian and Discord make the work easier to understand, but neither replaces the formal experiment record.https://irisseaman.com/wp-content/uploads/2026/10/twin_forge_current_architecture.pngThe current Twin Forge research architectureThe agents exchange specifications and evidence through the repository. Obsidian and Discord make the work easier to understand, but neither replaces the formal experiment record.https://irisseaman.com/wp-content/uploads/2026/10/twin_forge_two_agent_loop.pngThe first Twin Forge research loopThe first loop returned each result to the Scientist for a new question or a stop decision.2026-10-09T19:52:14+00:00monthlyhttps://irisseaman.com/resume/https://irisseaman.com/wp-content/uploads/2023/03/seaman_iris_resume.pngseaman_iris_resumehttps://irisseaman.com/wp-content/uploads/2023/03/seaman_iris_resume.jpgseaman_iris_resume2026-10-09T18:20:43+00:00weekly0.6https://irisseaman.com/2026/10/09/making-the-twin-forge-observable/2026-10-09T18:02:34+00:00monthlyhttps://irisseaman.com/2026/09/30/event-driven-python-reliable-ai-workflows/2026-10-10T21:36:50+00:00monthlyhttps://irisseaman.com/ml-algorithms/2026-09-29T18:54:40+00:00weekly0.6https://irisseaman.com/2026/09/29/production-llm-application-demo-to-production/2026-09-29T18:38:38+00:00monthlyhttps://irisseaman.com/homepage/https://irisseaman.com/wp-content/uploads/2026/09/ispy2_two_patient_rollout_static-2.pngispy2_two_patient_rollout_statichttps://irisseaman.com/wp-content/uploads/2026/09/crystal_edge_schematic-1.pngcrystal_edge_schematichttps://irisseaman.com/wp-content/uploads/2026/09/ssl_embedding_model_design-1.jpgssl_embedding_model_designhttps://irisseaman.com/wp-content/uploads/2026/09/endpoint_calibrated_spatial_forecasting-1.jpgendpoint_calibrated_spatial_forecastinghttps://irisseaman.com/wp-content/uploads/2026/09/endpoint_calibrated_spatial_forecasting-1.pngendpoint_calibrated_spatial_forecastinghttps://irisseaman.com/wp-content/uploads/2026/09/ispy2_two_patient_rollout_static-1-edited.pngispy2_two_patient_rollout_statichttps://irisseaman.com/wp-content/uploads/2026/09/ispy2_two_patient_rollout_static-1.pngispy2_two_patient_rollout_statichttps://irisseaman.com/wp-content/uploads/2026/09/graph_on_mri_radiologist_view-1.pnggraph_on_mri_radiologist_viewhttps://irisseaman.com/wp-content/uploads/2026/09/graph_on_mri_radiologist_view-5.jpggraph_on_mri_radiologist_viewhttps://irisseaman.com/wp-content/uploads/2026/09/img_4665-copy-2.jpgIMG_4665 copy2026-09-27T23:48:55+00:00weekly0.6https://irisseaman.com/2026/06/04/particle-filters/https://irisseaman.com/wp-content/uploads/2026/06/download.pngdownload2026-06-04T20:44:02+00:00monthlyhttps://irisseaman.com/2026/06/04/kalman-filters/2026-06-04T20:02:21+00:00monthlyhttps://irisseaman.com/2023/03/30/just-got-certified-as-an-mlexpert/2023-03-30T19:11:48+00:00monthlyhttps://irisseaman.com/2019/05/16/graphical-model-and-how-easy-it-is-to-break-conditional-independence/2023-03-26T20:00:03+00:00monthlyhttps://irisseaman.com/2019/06/10/sequential-monte-carlo-smc/https://irisseaman.com/wp-content/uploads/2019/06/1-2-1.png1 22024-02-02T01:27:10+00:00monthlyhttps://irisseaman.com/2019/06/07/annealing-importance-sampling/https://irisseaman.com/wp-content/uploads/2019/06/screen-shot-2019-06-07-at-1.33.26-pm.pngScreen Shot 2019-06-07 at 1.33.26 PM2019-06-11T01:45:40+00:00monthlyhttps://irisseaman.com/2019/06/01/mcmc-metropolis-hastings/2019-06-08T19:23:35+00:00monthlyhttps://irisseaman.com/2019/05/28/what-is-mcmc/https://irisseaman.com/wp-content/uploads/2019/06/screen-shot-2019-06-06-at-7.44.58-pm-2.pngScreen Shot 2019-06-06 at 7.44.58 PM 2https://irisseaman.com/wp-content/uploads/2019/06/screen-shot-2019-06-06-at-7.44.58-pm.pngScreen Shot 2019-06-06 at 7.44.58 PM2019-06-08T19:08:23+00:00monthlyhttps://irisseaman.com/2019/05/23/importance-sampling-vs-self-normalizing-is/2019-06-08T18:51:03+00:00monthlyhttps://irisseaman.comdaily1.02026-10-10T22:53:21+00:00