<![CDATA[AI+X Global Talent Community]]> https://gtc.blendedlearn.org Bettermode RSS Generator Tue, 28 Jul 2026 10:54:50 GMT Tue, 28 Jul 2026 10:54:50 GMT 60 <![CDATA[Integro Labs, LLC]]> https://gtc.blendedlearn.org/industry-3ggvoxci/post/integro-labs-llc-ueL9hyMkMizoYqa https://gtc.blendedlearn.org/industry-3ggvoxci/post/integro-labs-llc-ueL9hyMkMizoYqa Wed, 22 Jul 2026 11:19:06 GMT Integro Labs turns teams' knowledge into adaptable, owned AI capability — automation and operations delivered through their own open tooling, run as a managed service or installed in a client's environment. The constant across everything they ship is AIWG, their open agentic framework, and Fortemi, their production semantic-memory and search service. Their philosophy is local-first and no lock-in: systems are built to be durable and owned, not dependent on a single model or cloud provider.

]]> <![CDATA[Teaching Robots to See: A Career in Computer Vision and Robotics]]> https://gtc.blendedlearn.org/events-meetups-8up4ldya/post/teaching-robots-to-see-a-career-in-computer-vision-and-robotics-hJHOLzS3hQukzoC https://gtc.blendedlearn.org/events-meetups-8up4ldya/post/teaching-robots-to-see-a-career-in-computer-vision-and-robotics-hJHOLzS3hQukzoC Fri, 10 Jul 2026 15:14:33 GMT Join us for a candid Q&A session where a computer vision engineer who has shipped perception systems for robots in homes, farms, and warehouses talks through how the field actually works day to day. Hear directly about the technical trade-offs, deployment surprises, and career decisions behind building vision systems that have to work outside a benchmark.

Speaker: Dan Shao, Principal Computer Vision Engineer

Dan Shao is a computer vision and robotics engineer with 15+ years of industrial experience spanning healthcare, smart home devices, factory automation, and agriculture. She has led computer vision and machine learning work at iRobot (vision systems for Roomba navigation and docking), Root AI (perception systems for produce-harvesting robots), and UBTECH Robotics (posture detection and navigation for healthcare robots), and spent over two years at Cognex Corporation building vision systems for warehouse logistics — including work serving Amazon Robotics directly on edge cases like damaged packaging and distorted barcodes. She now leads computer vision, machine learning, and AI at Zelig, where she applies the same human-body perception and generative modeling techniques — pose estimation, motion synthesis, identity preservation — that underpin next-generation robot perception and human-robot interaction. Her focus throughout has been applying computer vision to real deployment constraints: complex lighting, new sensor technology, and edge computing on resource-limited platforms.

What You'll Hear:

The Speaker's Journey & Expertise

  • How she moved from research into industrial computer vision, and the inflection points along the way

  • What changed moving between industries — healthcare robots, smart home devices, warehouse logistics, agriculture

  • How industrial deployment work differs from research and academic settings

Real-World Projects & Impact

  • Building the vision system behind Roomba's navigation and docking

  • Solving edge cases in warehouse package and barcode reading for a strategic partner in robotics logistics

  • Vision challenges in agricultural robotics — recognizing produce reliably under unpredictable field conditions

  • Applying generative modeling and human-body pose estimation to virtual try-on today — the same identity-preserving motion synthesis and posture modeling that power human-robot interaction and animation in robotics

  • What breaks when a computer vision system leaves the benchmark and enters the real world

The Technical Landscape (Through Her Lens)

  • Classical vision vs. learned methods — when each still wins

  • Working around difficult lighting, novel camera sensors, and constrained edge compute

  • Human posture detection and safe human-robot interaction

Career & Opportunity

  • What companies building robots actually look for in computer vision engineers

  • How to build project experience that demonstrates deployment thinking, not just benchmark accuracy

  • Paths forward in robotics, computer vision, and applied ML

Why This Matters

You're not just learning a field — you're learning from someone who has shipped it, across four very different industries. Deployment constraints don't show up in a textbook, and the judgment calls that separate a working system from a benchmark result are usually invisible until you've hit them yourself. This session connects that lived experience to the technical fundamentals: how real constraints shape engineering decisions, and where the opportunities in applied computer vision are opening up.

Who Should Attend

  • Students exploring computer vision or robotics and wondering what real deployment work looks like

  • Anyone curious about the Computer Vision and Robotics – Amazon Robotics PBL

  • Future perception engineers, robotics engineers, or applied ML practitioners

]]> <![CDATA[ESG-Integrated Portfolio Selection for Technology Companies: A TOPSIS-Based Multi-Criteria Decision-Making Approach]]> https://gtc.blendedlearn.org/project-team-matching-uuc2fiio/post/esg-integrated-portfolio-selection-for-technology-companies-a-topsis-J3CT4F36mO6fPZZ https://gtc.blendedlearn.org/project-team-matching-uuc2fiio/post/esg-integrated-portfolio-selection-for-technology-companies-a-topsis-J3CT4F36mO6fPZZ Fri, 10 Jul 2026 12:00:22 GMT This project addresses the challenge of ESG-integrated portfolio selection for technology companies, where investors must evaluate firms across competing priorities such as sustainability performance, financial fundamentals, market strength, risk exposure, and public controversy. Traditional portfolio selection mainly focuses on return and risk, while conventional ESG ratings are often fragmented, inconsistent, and difficult to compare across data providers. This makes it challenging to build a transparent and reliable company-level investment ranking framework.

To solve this problem, we develop an automated Azure cloud ETL pipeline that extracts, transforms, and standardizes multi-source data from SEC EDGAR filings, SEC XBRL, Alpha Vantage, World Bank, Kaggle, NewsAPI, and Yahoo Finance. The cleaned dataset is then used to construct ESG, financial, market, and risk indicators. An entropy-weighted TOPSIS model is applied to generate a data-driven company ranking, while a controversy penalty adjustment is added to capture negative news exposure and reputational downside risks that may not be reflected in conventional ESG scores.

Our project provides a practical and scalable decision-support framework for ESG-aware technology investment screening. It helps investors, asset managers, and sustainability analysts benchmark companies more transparently, identify firms with stronger ESG-financial balance, and account for controversy-related risks in portfolio decision-making. By combining automated data integration, objective entropy weighting, TOPSIS ranking, and controversy adjustment, the project demonstrates how multi-criteria decision-making can support real-world investment decisions involving tradeoffs among profitability, sustainability, risk, and corporate responsibility.

]]> <![CDATA[Strategic Fleet Allocation: Outsmarting The NYC Oligopoly With Adversarial Bandits]]> https://gtc.blendedlearn.org/project-team-matching-uuc2fiio/post/strategic-fleet-allocation-outsmarting-the-nyc-oligopoly-with-xyXgkkgZxF8yCst https://gtc.blendedlearn.org/project-team-matching-uuc2fiio/post/strategic-fleet-allocation-outsmarting-the-nyc-oligopoly-with-xyXgkkgZxF8yCst Fri, 10 Jul 2026 11:59:37 GMT  This project investigates how competing, reinforcement learning (RL)-based ride-hailing platforms (like Uber and Lyft) interact within the New York City duopoly/oligopoly market. Specifically, it explores algorithmic collusion—whether independent AI agents managing different platforms will autonomously learn to coordinate, artificially keeping surge prices high to exploit passengers without any explicit human agreement, or if they will fall into a competitive price war. We find that RL-based ride-hailing platforms converged to similar high-revenue levels instead of engaging in aggressive price competition. The response to fixed and adaptive competitors shows that RL agents can form a cooperative pricing equilibrium without explicit agreement.

]]> <![CDATA[ESG-Integrated Portfolio Selection for Technology Companies: A TOPSIS-Based Multi-Criteria Decision-Making Approach]]> https://gtc.blendedlearn.org/project-team-matching-uuc2fiio/post/esg-integrated-portfolio-selection-for-technology-companies-a-topsis-M8bMpQ3gGUp9qet https://gtc.blendedlearn.org/project-team-matching-uuc2fiio/post/esg-integrated-portfolio-selection-for-technology-companies-a-topsis-M8bMpQ3gGUp9qet Fri, 10 Jul 2026 11:57:29 GMT

This project addresses the challenge of ESG-integrated portfolio selection for technology companies, where investors must evaluate firms across competing priorities such as sustainability performance, financial fundamentals, market strength, risk exposure, and public controversy. Traditional portfolio selection mainly focuses on return and risk, while conventional ESG ratings are often fragmented, inconsistent, and difficult to compare across data providers. This makes it challenging to build a transparent and reliable company-level investment ranking framework.

To solve this problem, we develop an automated Azure cloud ETL pipeline that extracts, transforms, and standardizes multi-source data from SEC EDGAR filings, SEC XBRL, Alpha Vantage, World Bank, Kaggle, NewsAPI, and Yahoo Finance. The cleaned dataset is then used to construct ESG, financial, market, and risk indicators. An entropy-weighted TOPSIS model is applied to generate a data-driven company ranking, while a controversy penalty adjustment is added to capture negative news exposure and reputational downside risks that may not be reflected in conventional ESG scores.

Our project provides a practical and scalable decision-support framework for ESG-aware technology investment screening. It helps investors, asset managers, and sustainability analysts benchmark companies more transparently, identify firms with stronger ESG-financial balance, and account for controversy-related risks in portfolio decision-making. By combining automated data integration, objective entropy weighting, TOPSIS ranking, and controversy adjustment, the project demonstrates how multi-criteria decision-making can support real-world investment decisions involving tradeoffs among profitability, sustainability, risk, and corporate responsibility.

]]> <![CDATA[ESG-Integrated Portfolio Selection for Technology Companies: A TOPSIS-Based Multi-Criteria Decision-Making Approach]]> https://gtc.blendedlearn.org/project-team-matching-uuc2fiio/post/esg-integrated-portfolio-selection-for-technology-companies-a-topsis-JDzx0WxL7SzY3oj https://gtc.blendedlearn.org/project-team-matching-uuc2fiio/post/esg-integrated-portfolio-selection-for-technology-companies-a-topsis-JDzx0WxL7SzY3oj Fri, 10 Jul 2026 11:56:16 GMT

This project addresses the challenge of ESG-integrated portfolio selection for technology companies, where investors must evaluate firms across competing priorities such as sustainability performance, financial fundamentals, market strength, risk exposure, and public controversy. Traditional portfolio selection mainly focuses on return and risk, while conventional ESG ratings are often fragmented, inconsistent, and difficult to compare across data providers. This makes it challenging to build a transparent and reliable company-level investment ranking framework.

To solve this problem, we develop an automated Azure cloud ETL pipeline that extracts, transforms, and standardizes multi-source data from SEC EDGAR filings, SEC XBRL, Alpha Vantage, World Bank, Kaggle, NewsAPI, and Yahoo Finance. The cleaned dataset is then used to construct ESG, financial, market, and risk indicators. An entropy-weighted TOPSIS model is applied to generate a data-driven company ranking, while a controversy penalty adjustment is added to capture negative news exposure and reputational downside risks that may not be reflected in conventional ESG scores.

Our project provides a practical and scalable decision-support framework for ESG-aware technology investment screening. It helps investors, asset managers, and sustainability analysts benchmark companies more transparently, identify firms with stronger ESG-financial balance, and account for controversy-related risks in portfolio decision-making. By combining automated data integration, objective entropy weighting, TOPSIS ranking, and controversy adjustment, the project demonstrates how multi-criteria decision-making can support real-world investment decisions involving tradeoffs among profitability, sustainability, risk, and corporate responsibility.

]]> <![CDATA[Prediction-Noise-Aware Post-hoc Fairness in Resource Allocation]]> https://gtc.blendedlearn.org/project-team-matching-uuc2fiio/post/prediction-noise-aware-post-hoc-fairness-in-resource-allocation-xxM18MlEgpCLC8p https://gtc.blendedlearn.org/project-team-matching-uuc2fiio/post/prediction-noise-aware-post-hoc-fairness-in-resource-allocation-xxM18MlEgpCLC8p Fri, 10 Jul 2026 11:44:09 GMT This project studies the robustness of GIFF, a post-hoc fairness method for multi-agent resource allocation, when the utility predictions used by the allocation rule are noisy. We first reproduce GIFF’s clean-data fairness advantage on the homelessness allocation pipeline, then stress-test the method by injecting prediction noise into the predicted utilities. We find that vanilla GIFF can lose its fairness advantage and even backfire in low-advantage groups under high noise. To address this, we propose gated-GIFF, a conservative reliability-aware wrapper that uses GIFF only when its estimated advantage over the baseline clears a safety margin, and otherwise falls back to the baseline. Our results suggest that post-hoc fairness methods should know when not to intervene.

]]> <![CDATA[Prediction-Noise-Aware Post-hoc Fairness in Resource Allocation]]> https://gtc.blendedlearn.org/project-team-matching-uuc2fiio/post/prediction-noise-aware-post-hoc-fairness-in-resource-allocation-7qj0higMI7YHSOu https://gtc.blendedlearn.org/project-team-matching-uuc2fiio/post/prediction-noise-aware-post-hoc-fairness-in-resource-allocation-7qj0higMI7YHSOu Fri, 10 Jul 2026 11:35:56 GMT This project studies the robustness of GIFF, a post-hoc fairness method for multi-agent resource allocation, when the utility predictions used by the allocation rule are noisy. We first reproduce GIFF’s clean-data fairness advantage on the homelessness allocation pipeline, then stress-test the method by injecting prediction noise into the predicted utilities. We find that vanilla GIFF can lose its fairness advantage and even backfire in low-advantage groups under high noise. To address this, we propose gated-GIFF, a conservative reliability-aware wrapper that uses GIFF only when its estimated advantage over the baseline clears a safety margin, and otherwise falls back to the baseline. Our results suggest that post-hoc fairness methods should know when not to intervene.

]]> <![CDATA[Prediction-Noise-Aware Post-hoc Fairness in Resource Allocation]]> https://gtc.blendedlearn.org/project-team-matching-uuc2fiio/post/prediction-noise-aware-post-hoc-fairness-in-resource-allocation-BsBkxcDRyIWRka8 https://gtc.blendedlearn.org/project-team-matching-uuc2fiio/post/prediction-noise-aware-post-hoc-fairness-in-resource-allocation-BsBkxcDRyIWRka8 Fri, 10 Jul 2026 11:31:49 GMT This project studies the robustness of GIFF, a post-hoc fairness method for multi-agent resource allocation, when the utility predictions used by the allocation rule are noisy. We first reproduce GIFF’s clean-data fairness advantage on the homelessness allocation pipeline, then stress-test the method by injecting prediction noise into the predicted utilities. We find that vanilla GIFF can lose its fairness advantage and even backfire in low-advantage groups under high noise. To address this, we propose gated-GIFF, a conservative reliability-aware wrapper that uses GIFF only when its estimated advantage over the baseline clears a safety margin, and otherwise falls back to the baseline. Our results suggest that post-hoc fairness methods should know when not to intervene.

]]> <![CDATA[Prediction-Noise-Aware Post-hoc Fairness in Resource Allocation]]> https://gtc.blendedlearn.org/project-team-matching-uuc2fiio/post/prediction-noise-aware-post-hoc-fairness-in-resource-allocation-9RjdcCRhHFTIGse https://gtc.blendedlearn.org/project-team-matching-uuc2fiio/post/prediction-noise-aware-post-hoc-fairness-in-resource-allocation-9RjdcCRhHFTIGse Fri, 10 Jul 2026 11:29:32 GMT This project studies the robustness of GIFF, a post-hoc fairness method for multi-agent resource allocation, when the utility predictions used by the allocation rule are noisy. We first reproduce GIFF’s clean-data fairness advantage on the homelessness allocation pipeline, then stress-test the method by injecting prediction noise into the predicted utilities. We find that vanilla GIFF can lose its fairness advantage and even backfire in low-advantage groups under high noise. To address this, we propose gated-GIFF, a conservative reliability-aware wrapper that uses GIFF only when its estimated advantage over the baseline clears a safety margin, and otherwise falls back to the baseline. Our results suggest that post-hoc fairness methods should know when not to intervene.

]]>