MUMBAI, India, Aug. 24 -- Intellectual Property India has published a patent application (202641098448 A) filed by Santhiram Engineering College on August 14, 2026, for Double Dqn-Powered Intelligent Framework For Optimizing Student-Alumni Engagement And Career Mentorship.
Inventors include M Sharmila Devi; E Haritha; K Sai Prasad Reddy; Gurakala Narasimha; Challa Tharaka Rajesh; Dasari Gangadhara Sai; Chakali Sunil Kumar; Mopuri Geetha; Ramireddygari Laxmi Harikaa; Mekala Radhika; and Mangali Pushpa Latha.
The application for the patent was published on August 21, 2026, under issue no. 34/2026.
Abstract: A computer-implemented framework and method for optimizing student-alumni engagement and career mentorship employs a Double Deep Q-Network (Double DQN) reinforcement-learning agent comprising an online neural network and a periodically synchronized target neural network, wherein action selection is decoupled from action evaluation so as to materially reduce the Q-value overestimation bias inherent in conventional single-network DQN approaches. Student profile, academic, and activity data, together with alumni profile, industry, and availability data, are ingested, preprocessed, and encoded by a state representation encoder into a joint fixed- length state vector representing the current engagement context. The Double DQN engagement engine processes this state vector through its online network, which selects the mentorship-related action associated with the highest estimated action-value, while a structurally identical but independently synchronized target network evaluates that selected action to produce a stable temporal-difference training target. An experience replay buffer stores state-action-reward-next-state transitions arising from executed recommendations and supplies randomly sampled minibatches for training, thereby improving sample efficiency and breaking harmful temporal correlations between consecutive interactions. A reward computation module calculates a composite scalar reward as a weighted combination of an engagement sub- reward (e.g., session completion, response rate, recency), a career-outcome sub-reward (e.g., internship conversion, placement, sustained mentorship duration), and a penalty sub-term (e.g., no-shows, mismatches, mentor overload). An action-selection and mentorship-matching module translates the resulting policy output into a concrete recommendation, such as a specific alumni-mentor pairing, a suggested engagement activity, or a scheduling nudge, which is delivered to students and alumni through an engagement interface. Outcome feedback captured through that interface is fed back as the next observed state and reward, closing the reinforcement- learning loop. Through repeated interaction cycles, the resulting policy converges toward mentorship-matching and engagement-scheduling decisions that maximize cumulative long-term student career benefit and alumni engagement and retention, while remaining extensible to alternative embodiments including dueling network architectures, prioritized experience replay, and multi-campus or federated deployment configurations.
Disclaimer: Curated by HT Syndication.
