NSRI Journal + Archive
Original student research, reviewed and preserved in publicOpen accessNSRI Journal + Archive
Original student research, reviewed and preserved in publicOpen accessIssue index
NSRI Student Research Journal
Online ISSN: 3143-5653
This page provides ordinary HTML links to article records so readers and scholarly crawlers can discover each article URL directly.
Original Research | Journal Publication | NSRI-J-2026-0093
Authors: Aditya Chandran Arvind
Publication date: 2026-09-05
PDF: Open PDF
ISSN: Online ISSN: 3143-5653
Higher-order data are often compressed into pairwise co-occurrence counts, but this can hide how pairs are assembled into triples. We ask for the largest possible difference in deterministic cascade size between simple 3-uniform hypergraphs with the same labeled exact pair-codegree matrix. Starting from two active vertices, a triple activates its third vertex when its other two are active. We define M₂(n) as the maximum closure-size gap inside any projection fiber on n vertices. A direct proof gives M₂(n)=0 for 2≤n≤5 and M₂(n)=n−3 for n≥6. The upper bound follows because equal codegrees force both cascades either to stop at two or reach at least three vertices. A five-edge Pasch-trade-plus-context construction attains the bound and extends to every larger n. Fresh exhaustive enumeration of all 2²⁰ simple 3-uniform hypergraphs on six labeled vertices independently checks the first ambiguous order, a maximum gap of three, and edge thresholds: four edges for different final sets, five for different sizes, and seven when non-isomorphism is additionally required on six vertices. Two separately implemented censuses agree on 69 shared summary fields. The result is a precise warning about information lost under pairwise compression for one abstract deterministic closure rule; it is not an empirical claim about real epidemics or social networks.
Citation: Aditya Chandran Arvind (2026). Exact Cascade Uncertainty Under Pairwise Compression. NSRI Student Research Journal. 1(1). Article 0093. 10.67677/hwc30rbc.
Review Article | Journal Publication | NSRI-J-2026-0094
Background: Diabetic peripheral neuropathy (DPN) affects at least half of people with diabetes and is a major risk factor for foot ulceration and amputation. Because nerve damage is least reversible late, a screening method's value depends on whether it can be delivered widely and early, not accuracy. Methods: This narrative review compared four DPN detection methods, nerve conduction studies, clinical sensory tests, plantar pressure analysis, and wearable gait sensing, across cost, setup, time, scalability, and accuracy. Results: Accuracy and scalability traded off. Nerve conduction studies are the accepted reference standard for large-fiber function but too specialist-dependent for population screening, while the cheapest test, the 10-gram monofilament, suits advanced disease better than early disease. Sensor-based methods report high accuracy from small, internally validated samples. Conclusion: No method is simultaneously cheap, fast, accurate, and sensitive to early disease. The unmet need is a scalable test externally validated for early detection.
Citation: Aayush Shah (2026). CHEAP, FAST, OR ACCURATE: COMPARING THE PRACTICALITY OF SCREENING METHODS FOR DIABETIC PERIPHERAL NEUROPATHY. NSRI Student Research Journal. 1(1). Article 0094. 10.67677/phztct86.
Perspective or Conceptual Paper | Journal Publication | NSRI-J-2026-0095
Authors: Zohaib Bari, Keshav Ragunath, Suyash Tripathi
Publication date: 2026-09-05
PDF: Open PDF
ISSN: Online ISSN: 3143-5653
Background. Off-exchange venues executed a record 51.8% of United States equity share volume in January 2025. A widely repeated interpretation attributes this migration to machine learning embedded in matching and routing infrastructure. This paper tests whether public data can support that attribution, and reports what they support instead.\n\nMethods. Six datasets were compiled from FINRA, SEC, and ESMA publications, from enforcement releases, and from the peer-reviewed and working-paper literature, and are provided in full with a calculation workbook in which every derived statistic is a live formula. A three-tier evidence protocol separates directly measured values from author-derived arithmetic and from interpretive attribution, and ten robustness tests are reported alongside the results they test.\n\nResults. In share terms, non-ATS off-exchange venues account for 86.3% of the 14.6 percentage-point rise in off-exchange share since 2019, a result robust to endpoint and window variation but not to dollar denomination, under which the figure falls to 54.1%. The 15.93 basis-point price-impact differential between retail and institutional execution indicates lower adverse selection in retail flow and is not, as it is often read, an execution penalty borne by retail investors. Sixteen enforcement actions totaling $330.15 million cluster in 2015 to 2018, and none alleges conduct involving a machine-learning system.\n\nConclusion. The public record establishes a compositional shift toward bilateral internalization, a segmented market, and a disclosure regime that has not been updated for model-driven venues. It does not establish that machine learning caused any of these. The governance question stands without the attribution, and a difference-in-differences design using audit-trail microdata is specified that would settle it.
Citation: Zohaib Bari, Keshav Ragunath, Suyash Tripathi (2026). Off-Exchange Migration and the Limits of AI Attribution: A Data Synthesis of Venue Composition, Execution Quality, and Enforcement in U.S. Equity Markets, 2019 to 2025. NSRI Student Research Journal. 1(1). Article 0095. 10.67677/pa9umval.
Original Research | Journal Publication | NSRI-J-2026-0092
Background: Alzheimer's disease (AD) affects over 55 million people worldwide, and early detection through speech analysis offers a low-cost, non-invasive alternative to neuroimaging. However, clinical deployment of speech-based classifiers faces underexplored challenges: domain shifts across recording environments, limited labeled data at new sites, and a need for calibrated uncertainty estimates. Methods: We evaluated Random Forest and XGBoost classifiers on DementiaBank (295 Cookie Theft recordings from 292 unique speakers; 45 control, 250 AD). We extracted 1,589 features combining 53 hand-crafted acoustic descriptors (librosa, Praat) with wav2vec 2.0 and HuBERT embeddings, reduced to 128 dimensions via PCA. Robustness was assessed using 8 controlled acoustic simulations (microphone variation, background noise at 10-25 dB SNR, and reverberation at RT60 0.3-0.8 s) calibrated to published clinical measurements. Results: Both models achieved 81.4% raw accuracy (48/59; macro F1: 0.594; balanced accuracy: 0.581) on clean data, below the majority-class raw baseline of 84.7% but above it in balanced accuracy (0.500) and macro F1 (0.459). Under 7 of 8 controlled acoustic simulations, raw accuracy converged to 84.7%, matching the majority-class rate. A supplementary confusion-matrix analysis using a simplified verification model confirmed that this pattern reflects majority-class defaulting: control recall dropped to 0.000 under microphone and noise simulations. Random Forest achieved low calibration error (ECE = 0.042) on the 59-sample test set. In a single-draw few-shot experiment, XGBoost yielded 67.8% with K=10 examples per class. Conclusion: On clean data, acoustic features combined with self-supervised embeddings produce classifiers with meaningful balanced performance despite severe class imbalance (85% AD). The controlled acoustic simulation framework provides a reproducible evaluation methodology, though our results reveal that apparent robustness may reflect majority-class defaulting rather than genuine invariance, an important methodological caution for future work.
Citation: Aditya Amarjeet Singh (2026). Speech-Based Alzheimer's Detection Under Controlled Acoustic Perturbations: A Baseline Study of Class Imbalance and Robustness. NSRI Student Research Journal. 1(1). Article 0092. 10.67677/t4vh5i27.
Review Article | Journal Publication | NSRI-J-2026-0091
Authors: Sanat Tibrewal, Pranesh Nagarajan
Publication date: 2026-08-23
PDF: Open PDF
ISSN: Online ISSN: 3143-5653
Background: Consumer neuroscience has identified loss aversion and reward anticipation as two core mechanisms shaping purchase decisions, but traditional measurement methods such as fMRI and EEG remain expensive and difficult to scale. Methods: This narrative review synthesizes peer-reviewed literature on prospect theory, dopaminergic reward circuitry, consumer neuroscience, and recent work on large language model (LLM) based digital twins and silicon sampling to evaluate whether LLMs can approximate these mechanisms. Results: One recent consumer digital-twin study reported 86% future-purchase prediction accuracy using individualized fine-tuning on Amazon transaction data, and a large-scale behavioral benchmark found that even the best-performing LLMs achieved only modest simulation fidelity (40.80/100) across 20 diverse human-behavior datasets. Available studies have not established that LLMs replicate the behavioral asymmetry of loss aversion, the temporal dynamics of reward anticipation, or individual variation in these mechanisms. Conclusion: Current evidence suggests that LLMs function as useful behavioral prediction tools for consumer research but have not yet been shown to reproduce the mechanistic processes documented in consumer neuroscience.
Citation: Sanat Tibrewal, Pranesh Nagarajan (2026). Can LLM-Based Digital Twins Approximate Human Loss Aversion and Reward Anticipation in Consumer Decision-Making? A Narrative Review of Neural Benchmarks and Computational Limitations. NSRI Student Research Journal. 1(1). Article 0091. 10.67677/fp71k64a.
Methods or Technical Paper | Journal Publication | NSRI-J-2026-0090
DOSEFIELD is an open, deterministic one-dimensional model of the deep-space galactic cosmic ray (GCR) radiation environment: a parametric GCR spectrum is transported through a slab shield in the continuous-slowing-down approximation and converted to absorbed dose and ICRP-60 dose-equivalent in water. Every constant is taken from a cited primary source, and no parameter is fitted to any validation target. Against NIST PSTAR proton stopping-power tables the Bethe-Bloch engine agrees to a maximum deviation of 1.55% above 10 MeV across five low-Z materials (4.03% including the 1-10 MeV region, where shell corrections are omitted). Against the MSL/RAD cruise measurement, evaluated at a solar modulation and shield thickness chosen independently of the measurement, the model returns an absorbed dose of 0.308 mGy/day against a measured 0.458 +/- 0.032 mGy/day (ratio 0.67x), a dose-equivalent of 1.473 mSv/day against a measured 1.75 +/- 0.30 mSv/day (ratio 0.84x), and a mean quality factor <Q> of 4.78 against a measured 3.82 +/- 0.25 (ratio 1.25x). The propagated 1-sigma-style input uncertainty on the dose-equivalent is 14.6% (GCR flux, stopping-power data and this run's PSTAR deviation combined in quadrature), which does not cover the absorbed-dose shortfall: that shortfall is model-form error from un-transported secondary production, and it is reported rather than corrected. Contemporaneous in-cruise neutron measurements (6 +/- 2 microGy/day) account for only about 4% of the model's absorbed-dose deficit, so the deficit is attributed to secondary charged particles and target fragments, not neutrons. At equal areal density the model ranks five shield materials strictly by hydrogen content (H2 < CH4 < polyethylene < water < aluminium) at every thickness tested from 5 to 40 g/cm2; liquid hydrogen reduces dose-equivalent by up to 44.2% relative to aluminium. This paper reports a methods and validation study, not a new physical result: the value of the work is a fully inspectable, source-traceable implementation and an honest, quantitative account of where and why it disagrees with flight data.
Citation: Zhanbolat Izbanov (2026). DOSEFIELD: A Transparent 1-D Deep-Space Radiation Dose and Shielding Model with Source-Traceable Constants. NSRI Student Research Journal. 1(1). Article 0090. 10.67677/p3bzqdbb.
Review Article | Journal Publication | NSRI-J-2026-0089
Severe hantavirus infection produces extensive vascular leakage, driven by a combination of endothelial dysfunction, dysregulated inflammation, VEGF-mediated hyperpermeability, angiopoietin imbalance, and degradation of the endothelial glycocalyx. This narrative review draws on the peer-reviewed literature, with particular attention to work published between 2016 and 2026, to examine how hantavirus-induced vascular permeability arises in hemorrhagic fever with renal syndrome (HFRS) and hantavirus pulmonary syndrome (HPS), and how Angiopoietin-1 (Ang-1)/Tie2 signaling and glycocalyx injury contribute to endothelial barrier failure. Direct hantavirus evidence shows that Ang-1 inhibits hantavirus-directed endothelial permeability in vitro and that angiopoietin dysregulation tracks with disease severity in patients; findings that Tie2 activation curbs vascular leakage in sepsis and other non-hantavirus models are mechanistically suggestive but remain inferential rather than hantavirus-specific. No hantavirus animal model or clinical trial has yet evaluated Ang-1/Tie2-directed therapy, so this review treats the approach as a testable hypothesis rather than an established or recommended treatment; confirming its therapeutic value will require dedicated hantavirus-specific animal and clinical studies.
Citation: Atharv Dhore (2026). Angiopoietin-1/Tie2 Activation as a Therapeutic Hypothesis for Hantavirus-Induced Vascular Permeability. NSRI Student Research Journal. 1(1). Article 0089. 10.67677/vlv4o75m.
Review Article | Journal Publication | NSRI-J-2026-0001
Study objective: To investigate how school communities can support long-term healthy lifestyle habits among adolescents in high-stress academic environments through a theory-based intervention design. Design: Narrative literature review and theory-based intervention design. Setting: Secondary school and community environments. Participants: Adolescents aged 12–18 years in high-pressure academic settings. Proposed Intervention: We propose the Daily 3 Initiative, a school-community programme embedding three daily micro-habits (hydration, light movement, and focused breathing) into existing morning homeroom periods, supported by a peer Habit Buddy system and extended to family settings. Key findings from the literature: Four findings were synthesised: (1) chronic academic stress is associated with deterioration in adolescent lifestyle behaviours; (2) school-based interventions with limited structured contact points often show limited sustained behavioural change; (3) micro-habits anchored to existing routines demonstrate stronger long-term retention; and (4) peer accountability structures are associated with improved habit maintenance in adolescents. Conclusions: Drawing on the above findings, we propose the Daily 3 Initiative as a theory-based, zero-cost design. The intervention has not yet been tested as a unified programme. Future research should include feasibility studies and controlled trials.
Citation: Kamrul Hassan (2026). Embedding micro-habits in school routines: how communities can help young people build sustainable healthy lifestyles. NSRI Student Research Journal. 1(1). Article 0001. 10.67677/hltx1ari.
Case Study or Report | Journal Publication | NSRI-J-2026-0088
This study examines maternal healthcare coordination issues in rural Khyber Pakhtunkhwa (KPK), Pakistan. Despite current infrastructure, rural communities face significant obstacles due to the uneven geographic distribution of physicians and fragmented health governance. The study suggests a decentralized digital approach that equips Lady Health Workers (LHWs) with AI-assisted triage and routing, alongside a telemedicine welfare fund. Analysis of existing digital systems and LHW programs suggests that bridging the gap between discovery and care via trusted community intermediaries may significantly lower preventable maternal mortality without the need for additional physical facilities. Findings suggest that the healthcare crisis in KPK appears to be primarily a problem of coordination and discovery rather than a shortage of infrastructure.
Citation: Muhammad Nameer Shah (2026). Empowering Lady Health Workers with AI Routing to Solve Healthcare Coordination Failures in Rural Khyber Pakhtunkhwa. NSRI Student Research Journal. 1(1). Article 0088. 10.67677/k7x4m2qp.
Original Research | Journal Publication | NSRI-J-2026-0064
Background: Chronic inflammation and metabolic disorders, including cardiovascular disease and type 2 diabetes mellitus, share a common pathophysiological basis in dysregulated lipid mediator signaling. Soluble epoxide hydrolase (sEH; EC 3.3.2.10) and cyclooxygenase-2 (COX-2) are validated pharmacological targets in these pathways. Current synthetic sEH inhibitors face challenges including metabolic instability and limited selectivity, motivating exploration of natural product-derived scaffolds. Phytohormones represent an underexplored ligand space for multi-target anti-inflammatory drug discovery. Objective: This study aimed to identify phytohormone-derived compounds with favorable predicted binding affinity at the sEH catalytic site (PDB: 3EBL) and COX-2 (PDB: 5F19), and to assess their oral drug-likeness through ADME profiling. Methods: Seven structurally diverse phytohormones were prepared using Open Babel v2.4.1 and docked against sEH (PDB: 3EBL) and COX-2 (PDB: 5F19) using AutoDock Vina v1.1.2. Grid boxes were centered on the co-crystallized ligand binding site. Docking protocol validity was confirmed by self-docking (RMSD < 2.0 Å). ADME properties and Lipinski drug-likeness were evaluated using SwissADME. Key Results: Abscisic acid (ABA) showed the most favorable predicted binding affinity at the sEH active site (ΔG = −6.4 kcal/mol), with predicted hydrogen-bond interactions at residues ASP213 and ARG210, consistent with known sEH pharmacophoric requirements. At COX-2, ABA also produced the lowest predicted energy (−5.6 kcal/mol). ABA, jasmonic acid, methyl jasmonate, and indole-3-acetic acid satisfied Lipinski’s Rule of Five, with ABA displaying high predicted gastrointestinal absorption (TPSA = 77.8 Ų) and no structural alerts. Conclusion: Abscisic acid is identified as a computationally prioritized lead for sEH inhibition with an acceptable in silico pharmacokinetic profile. Experimental validation through enzyme inhibition assays and biophysical binding studies is required before pharmacological conclusions can be drawn.
Citation: Mohd Sameer (2026). Molecular Docking-Based Screening of Phytohormone-Derived Compounds Against Soluble Epoxide Hydrolase and Cyclooxygenase-2: An In Silico Study With ADME-Based Pharmacokinetic Profiling. NSRI Student Research Journal. 1(1). Article 0064. 10.67677/m942rt1g.
Original Research | Journal Publication | NSRI-J-2026-0054
This paper examines whether Kazakhstan's Emissions Trading System (KazETS), introduced in 2013, generated a measurable causal reduction in the market valuation of high-emission industrial firms. Using an unbalanced panel of 52 publicly listed firms observed over 2008 to 2022, a difference-in-differences framework exploits cross-sectional variation in pre-policy carbon intensity to identify the effect on Tobin's Q. High-emission firms experienced a statistically significant decline of approximately 0.39 points in Tobin's Q relative to low-emission peers following policy implementation, a finding robust to firm and year fixed effects, industry-by-year interactions, staggered DiD estimators, propensity score matching, and placebo tests. The valuation discount is largest in mining and energy and is amplified for firms with higher institutional ownership. The results provide the first rigorous evidence that financial markets price regulatory climate risk in an emerging Central Asian economy, with direct implications for investors, corporate managers, and policymakers. Keywords: carbon risk, emissions trading system, firm valuation, Tobin's Q, difference-in-differences, Kazakhstan, emerging market
Citation: Yerdaulet Kenges (2026). Carbon Risk and Firm Valuation in an Emerging Market Emissions Trading System: Evidence from Kazakhstan. NSRI Student Research Journal. 1(1). Article 0054. 10.67677/tjct4yvi.
Review Article | Journal Publication | NSRI-J-2026-0052
This study investigated whether reversible, tissue-specific CRISPR circuits offer superior safety profiles compared with permanent knockout approaches in autoimmune disease gene therapy. CRISPR-based therapies have shown significant therapeutic potential in treating autoimmune disorders. Still, permanent gene editing introduces safety concerns, including off-target mutations, immune responses, unintended immunological vulnerabilities, and the inability to reverse harmful edits once treatment has occurred. The purpose of this research was to evaluate whether emerging reversible and tissue-specific CRISPR systems could address these limitations and provide a safer alternative. A structured secondary research methodology was used to analyze ten peer-reviewed studies published between 2016 and 2025. Studies were selected based on their relevance to CRISPR safety, autoimmune disease applications, reversibility, and tissue-specific control mechanisms. Evidence from in vitro experiments, animal studies, and human clinical trials was compared using five safety dimensions: off-target DNA modification, immune recognition, secondary immune vulnerabilities, reversibility, and tissue-targeting precision. The analysis found that permanent CRISPR systems exhibited the greatest safety risks. Studies documented large unintended DNA insertions, genomic instability, immune recognition of Cas9, and secondary vulnerabilities such as increased natural killer cell targeting following MHC knockout. Current autoimmune applications demonstrated strong therapeutic potential, including clinical remission in patients treated with CRISPR-edited CAR-T cells, but remained limited by their irreversible nature. Reversible and tissue-specific systems showed advantages in four major safety categories. Drug-inducible CRISPR systems enabled gene regulation to be activated and deactivated as needed, while microRNA-guided approaches achieved highly specific tissue targeting with extremely low off-target activity. Although these systems have not yet been tested in autoimmune clinical settings, available evidence suggests they provide substantially greater control over gene-editing activity. The findings support the conclusion that reversible, tissue-specific CRISPR circuits offer a promising pathway toward safer autoimmune gene therapies than permanent knockout approaches. However, current evidence remains largely limited to proof-of-concept and preclinical studies. Future research should focus on integrating reversible and tissue-specific control systems into a single therapeutic platform and evaluating their long-term safety and efficacy through autoimmune disease animal models and clinical trials.
Citation: Noma Ohonba (2026). An analysis of Reversible, Tissue-Specific CRISPR Circuits in gene editing safety compared to Permanent Knockout Approaches in Autoimmune Disease Gene Therapy. NSRI Student Research Journal. 1(1). Article 0052. 10.67677/g1hoki7k.
Original Research | Journal Publication | NSRI-J-2026-0045
Artificial intelligence and automation technologies are increasingly reshaping labor markets across advanced economies. This paper examines how AI adoption has influenced wage inequality and middle-skill employment across sixteen OECD countries between 2000 and 2024. The study investigates whether technological change has contributed to labor market polarization, and why some countries experienced significantly different inequality outcomes despite comparable levels of technological advancement. The research combines data from the OECD Income Distribution Database, the International Federation of Robotics, Oxford’s AI Readiness Index, and labor market datasets from the ILO, Eurostat, and the U.S. Bureau of Labor Statistics. To evaluate these relationships, the paper applies multiple econometric approaches, including pooled ordinary least squares, country fixed effects, instrumental variable estimation, and Arellano-Bond generalized method of moments models. Institutional variables such as union density, labor market policy spending, employment protections, and trade openness are also incorporated into the analysis. The findings reveal a strong and consistent relationship between higher AI adoption and rising income inequality. Countries with stronger AI readiness generally experienced larger increases in market-income Gini coefficients and sharper declines in middle-skill employment. However, the results also show substantial cross-country variation. Economies such as Sweden, Denmark, and Germany maintained relatively lower inequality growth despite widespread AI adoption, suggesting that labor institutions, coordinated wage bargaining systems, and workforce retraining policies significantly influence how technological gains are distributed. The paper argues that the economic effects of AI are shaped not only by technological progress itself, but also by the institutional structures surrounding labor markets. It concludes by discussing policy implications related to worker retraining, collective bargaining, labor protections, and long-term economic inclusion in increasingly automated economies.
Citation: Om Raut (2026). Machines, Markets, and the Disappearing Middle: AI and the Divided Job Market. NSRI Student Research Journal. 1(1). Article 0045. 10.67677/m25696bf.
Original Research | Journal Publication | NSRI-J-2026-0041
Scam and phishing detection systems typically rely on rigid heuristic rules or opaque large language models. This paper presents ScamShield: a hybrid, interpretable pipeline combining a 24-feature supervised machine learning model, an LLM semantic safety layer, and rule-based heuristics in a unified ensemble. Evaluated on a synthetic benchmark of 19,992 messages spanning 17 scam categories and validated on the UCI SMS Spam Collection (5,574 real-world messages), ScamShield achieves cross-validated F1 = 0.9969 ± 0.0004 on the synthetic benchmark and F1 = 0.9303 ± 0.0098 on real-world data — competitive with fine-tuned DistilBERT (estimated F1 ≈ 0.97–0.99) while requiring 125× less storage, operating at sub-5 ms inference latency, and providing full per-prediction interpretability via coefficient attribution. Statistical significance is confirmed by McNemar's test against all four baselines (p < 0.001). Adversarial evaluation reveals recall drops of 71–82% under obfuscation attacks, identifying semantic embeddings and character-level robustness as the primary improvement directions.
Citation: Vishwajeet Adkine (2026). ScamShield: Interpretable Multi-Signal Scam Detection. NSRI Student Research Journal. 1(1). Article 0041. 10.67677/4bcdmk3c.
Original Research | Journal Publication | NSRI-J-2026-0040
Alpha decay ejects MeV-scale alpha particles at ∼10^7 m/s, implying an extremely high effective exhaust velocity and specific impulse. This paper studies whether that exhaust can be used as a practical thrust source for deep-space missions without external electrical power by directly allowing decay products to leave a spacecraft in a preferred direction. The key obstacle is not exhaust velocity but momentum directionality and energy deposition: alpha emission is nearly isotropic and alpha particles have very short ranges in solids, so most decay energy becomes heat unless the radioactive material is structured as ultra-thin films. We develop a thin-film “decay sail” model, implement a reproducible Monte Carlo simulation (no external dependencies) to estimate an effective momentum coupling coefficient η(d/R), and perform isotope scaling studies for Po-210, Th-228, Pu-238, and Am-241. For a representative geometry d/R≈0.5 and an areal mass of 10 mg/cm², our model predicts an initial thrust density of ∼3.2×10^(-4) N/m² for Po-210 (tens of nN/cm²), consistent with published MCNP-based estimates for similar configurations. The same geometry implies kilowatt-to-tens-of-kilowatt per m² heat fluxes for short-half-life isotopes and mA/m²-scale charging currents unless charge neutralization is provided. We conclude that alpha-decay propulsion is generally thrust-poor but power-rich, best suited to niche missions that can tolerate extremely low thrust, very large emitter area, and significant thermal/charging engineering.
Citation: Ankush Dutta (2026). Alpha-Decay Alpha-Particle Propulsion for Deep-Space Missions: A First-Order Feasibility Study. NSRI Student Research Journal. 1(1). Article 0040. 10.67677/kajinkam.
Original Research | Journal Publication | NSRI-J-2026-0036
While industry and fossil fuels have greatly contributed to greenhouse gas emissions, the agricultural sector is another major driver, accounting for roughly 21-37% of human-related emissions (Mbow et al., 2022). The purpose of this study was to examine how soil pH influenced CO2 emissions and carbon and nitrogen retention in microbially activated biochar-amended soils. It was hypothesized that an alkaline pH would demonstrate greater carbon and nitrogen retention and lower cumulative CO2 emissions (ppm) over 21 days as opposed to acidic or neutral soils, because alkaline conditions are expected to favor microbial colonization of biochar pores and reduced carbon mineralization rates (Wang et al., 2023). To test this hypothesis, biochar-inoculum mixture was added to soil-filled jars divided into various pH levels. CO2 emissions were measured on days 1, 4, 7, 14, and 21. After 21 days, total organic carbon (TOC) and total organic nitrogen (TN) were measured with a soil testing kit. The ANOVA and post-hoc Tukey results supported the hypothesis: the alkaline group had the lowest CO2 emissions, while the neutral group had the highest (F(3, 116) = 42.01, p < .001). TOC and TN values showed that alkaline conditions experienced the highest nutrient retention (χ2(9, 120) = 72.37, p < .001). There is sufficient evidence to suggest that there is a difference in CO2 emissions based on soil pH. The results propose that microbially-activated biochar be instituted in agricultural settings to abate CO2 emissions and improve carbon and nitrogen retention. Future research should analyze enzyme and microbial activity within the soil and utilize specific microbial genus (e.g. Rhizobium spp. or Azospirillum spp.) instead of a general inocula.
Citation: Jay Shah (2026). The Impact of Soil pH on the Carbon and Nitrogen Retention Capacity and Cumulative CO2 Emissions in Microbially Activated Biochar-Amended Soil. NSRI Student Research Journal. 1(1). Article 0036. 10.67677/27fk9iyc.
Original Research | Journal Publication | NSRI-J-2026-0032
Fine-tuning pre-trained language models alter both performance metrics and internal representations, yet conventional loss measurements often fail to capture subtle shifts in high-dimensional embedding space. In this study, we introduce embedding drift, a scalar metric defined as the mean cosine distance between hidden-state vectors of fixed probe sentences, to quantify representational change during training. Each sentence is mapped to a 768-dimensional vector via mean pooling over token embeddings, and drift is computed as "drift"=1/N ∑_(i=1)^N▒〖(1-〗 (e_i^((0) )⋅e_i^((t) ))/(∥e_i^((0) )∥∥e_i^((t) )∥)) where N is the number of probe sentences, e_i^((0) )the initial embedding, and e_i^((t) )the embedding at step t. We performed two controlled experiments with DistilBERT on a subset of IMDb: a baseline (learning rate 5e-5) and a high learning rate (5e-4). Drift increased steadily from ~0.03 to ~0.26 in the baseline, whereas the high learning rate induced a rapid jump from ~0.22 to ~0.76, despite training loss showing minimal change. These results demonstrate that embedding drift provides a quantitative, vector-space measure of representational instability that conventional loss metrics may overlook, offering insight into internal model dynamics during fine-tuning.
Citation: Harshit Chaturvedy (2026). Embedding Drift as a Sensitive Indicator of Internal Representation Instability in Language Model Fine-Tuning. NSRI Student Research Journal. 1(1). Article 0032. 10.67677/owwzt9hr.
Original Research | Journal Publication | NSRI-J-2026-0031
Authors: Peter Olanrewaju, I. A. Abiade, J. A. Olaniran
Publication date: 2026-04-28
PDF: Open PDF
ISSN: Online ISSN: 3143-5653
Post-harvest losses of tomato fruits present a significant challenge in agricultural production, especially in areas lacking reliable electricity. This research developed and assessed a solar-powered cold storage system designed to extend tomato shelf life. The system featured a 4-liter insulated chamber with two Peltier modules, heat sinks, CPU fans for air circulation, a 12V-12Ah battery, a 100W solar panel, a charge controller, and temperature control. Design calculations indicated a cooling power requirement of 23W with a total system load of 158W. Performance evaluation indicated 67.71% cooling efficiency, decreasing chamber temperature from 31°C to 21.6°C within 60 minutes under no-load conditions and maintaining 23.3 to 24.5°C during loaded operation. Comparative storage trials conducted over 12 days showed substantial differences between cold-stored and ambient-stored tomatoes. Cold storage prevented physiological weight loss to 3.39% compared with 26.73% for ambient storage, maintained firmness at 163.67 N/m² compared to 207.33 N/m² for controls, preserved moisture content at 11.41% against 10.07%, and minimized color changes. Both conditions maintained stable pH values (4.1 for cold storage versus 4.0 for ambient). The research confirms that solar-powered cold storage offers a practical, sustainable solution for minimizing tomato losses in regions with inadequate grid electricity, substantially extending shelf life while preserving marketability qualities.
Citation: Peter Olanrewaju, I. A. Abiade, J. A. Olaniran (2026). Development of a Solar-powered Cold Storage for Tomato Fruits Preservation.. NSRI Student Research Journal. 1(1). Article 0031. 10.67677/ucx6e5uu.