BeveridgePatent, projects and publications
Dr. Sathiskumar JOTHI: AI and data engineering, computational and multiscale modelling of materials, hydrogen embrittlement, additive manufacturing, machining and composites.
Open-source release: Beveridge
October 2026. Beveridge is a family of System One decision models: give it a document and typed questions (yes/no, choice, score) and it returns each answer with a calibrated probability, in one pass, without generating text. Released as open weights under Apache-2.0, commercial use allowed.
The same queue, three models
16 items: 6 support messages, 5 requests to an AI assistant, 5 actions an AI agent wants to take. Same screen, same on-device OCR, same questions and the same rule: below 0.8 confidence the item goes to a person.
| Model | Where it ran | Automated | To a person | Wrong clicks | Median per item |
|---|---|---|---|---|---|
| Beveridge-Large-C | on this laptop (Apple M4 Pro GPU) | 10 | 6 | 0 | 116 ms |
| Beveridge-Pro-C | on one NVIDIA H200 GPU | 13 | 3 | 0 | 96 ms |
| Jev | cloud API, including the network | 13 | 3 | 0 | 132 ms |
Automated = items where the model clicked the final button itself. To a person = confidence below 0.8, or an agent action that deletes data, moves money or messages someone. Wrong = an automated click that differs from the correct one (example: allowing a payment). Median per item = the model's own time for one item's questions; for Jev it includes the network. Example: Beveridge-Large-C sent "Charged twice for order 88412" to a person at confidence 0.788 < 0.8, so it counts as to a person, not as wrong. A demo of 16 items, not a benchmark: the benchmarks are on the Beveridge page.
Beveridge-Pro-C-Web: a browser agent that decides every step
New: Beveridge-Pro-C with web browsing added. It picks the next operation and the element on the page, with a probability for each option, and keeps Beveridge-Pro-C's results on documents. Next step on web pages from site types it never saw: 98.2% (Beveridge-Pro-C: 49.4%). Live in the open-source browser loop jev-ultrafast: hotel search 3/3, Wikipedia 3/3. Keep a person in the loop for anything that buys, sends, deletes or signs.
Beveridge-Pro-C-Web
Hotel search: Lisbon, two filters, open the right hotel in a median 1.86 s. Wikipedia: find and open an article in a median 4.88 s. About 0.2 to 0.7 s per decision on one NVIDIA H200. Beveridge-Bench on documents: 92.1%.
Beveridge Voice: speak, Beveridge decides, the reply is spoken
Same model, different questions: each use case asks its own typed questions. Beveridge can make mistakes; keep a person in the loop for important decisions.
Beveridge-Large-C
Automates 76% of decisions at 95% accuracy on our held-out tests. About 5 ms per decision on one GPU, 60 ms on a laptop CPU.
Beveridge-Pro-C
Automates 69% of decisions at 95% accuracy. About 22 ms per decision on one GPU. Pair it with Beveridge-Large-C in route mode.
Automated share at 95% = the largest share of answers, taken from the most confident down, whose accuracy is still at least 95%. Example: 76% means 760 of 1,000 decisions go through automatically and are at least 95% right; the other 240 go to a person.
Patent and research projects
System for distributed data processing using clustering
Inventors: S Jothi, A Ganguly, C Cane · 23 citations (Google Scholar)
MultiHy: Multiscale Modelling of Hydrogen Embrittlement
Budget €5.12 million (EU contribution €3.39 million). Swansea University partner team.
Project on CORDISCitations per year
From Google Scholar. h-index = the largest h such that h works have at least h citations each; i10-index = works with at least 10 citations.
Journal articles (30)
Sorted by citations. Titles link to the publisher (DOI) where available, otherwise to Google Scholar.
| Cites | Year | Title and source |
|---|---|---|
| 343 | 2018 | Influence of powder characteristics and Additive manufacturing process parameters on the microstructure and mechanical behaviour of Inconel 625 fabricated by Selective Laser … CP Sathiskumar Jothi · Additive manufacturing, 2018 |
| 153 | 2016 | An investigation of micro-mechanisms in hydrogen induced cracking in nickel-based superalloy 718 S Jothi, SV Merzlikin, TN Croft, J Andersson, SGR Brown · Journal of Alloys and Compounds 664, 664-681, 2016 |
| 111 | 2019 | Prediction of surface roughness and material removal rate in wire electrical discharge machining on aluminum based alloys/composites using Taguchi coupled Grey Relational … T Thankachan, KS Prakash, R Malini, S Ramu, P Sundararaj, ... · Applied Surface Science 472, 22-35, 2019 |
| 68 | 2019 | Influence of materials and machining parameters on WEDM of Al/AlCoCrFeNiMo0.5 MMC S Karthik, KS Prakash, PM Gopal, S Jothi · Materials and Manufacturing Processes 34 (7), 759-768, 2019 |
| 67 | 2017 | Artificial neural network to predict the degraded mechanical properties of metallic materials due to the presence of hydrogen T Thankachan, KS Prakash, C David Pleass, D Rammasamy, B Prabakaran, S Jothi · international journal of hydrogen energy 42 (47), 28612-28621, 2017 |
| 63 | 2014 | Influence of grain boundary misorientation on hydrogen embrittlement in bi-crystal nickel S Jothi, TN Croft, SGR Brown · International journal of hydrogen energy 39 (35), 20671-20688, 2014 |
| 61 | 2018 | Reduced graphene oxide wrapped hierarchical TiO2 nanorod composites for improved charge collection efficiency and carrier lifetime in dye sensitized solar cells MR Subramaniam, D Kumaresan, S Jothi, JD McGettrick, TM Watson · Applied Surface Science 428, 439-447, 2018 |
| 58 | 2018 | Optimization of the Configuration of Porous Bone Scaffolds made of Polyamide/Hydroxyapatite Composites using Selective Laser Sintering for Tissue Engineering Applications M Ramu, M Ananthasubramanian, T Kumaresan, R Gandhinathan, S Jothi · Bio-Medical Materials and Engineering, DOI: 10.3233/BME-181020, 2018 |
| 58 | 2015 | Multi-phase modelling of intergranular hydrogen segregation/trapping for hydrogen embrittlement S Jothi, TN Croft, L Wright, A Turnbull, SGR Brown · International journal of hydrogen energy 40 (43), 15105-15123, 2015 |
| 53 | 2013 | Micromechanical modeling of piezo-magneto-thermo-elastic composite structures: Part I–Theory DA Hadjiloizi, AV Georgiades, AL Kalamkarov, S Jothi · European Journal of Mechanics-A/Solids 39, 298-312, 2013 |
| 52 | 2017 | Electrodeposition of r-GO/SiC nano-composites on magnesium and its corrosion behavior in aqueous electrolyte V Kavimani, K Soorya Prakash, R Rajesh, D Rammasamy, NB Selvaraj, T Yang, B Prabakaran, S Jothi · Applied Surface Science 424, 63-71, 2017 |
| 48 | 2012 | Micromechanical Modeling of Piezo-Magneto-Thermo-Elastic Composite Structures: Part II-Applications DA Hadjiloizi, AV Georgiades, AL Kalamkarov, S Jothi · European Journal of Mechanics, A/Solids 39, 313-327, 2012 |
| 45 | 2019 | Development and characterization of areca fiber reinforced polymer composite K Dinakaran, H Ramesh, AD Joseph, R Murugan, S Jothi · Materials Today: Proceedings 18, 934-940, 2019 |
| 44 | 2016 | Effects of additives on kinetics, morphologies and lead-sensing property of electrodeposited bismuth films AR Rajamani, S Jothi, MD Kumar, S Srikaanth, MK Singh, G Otero-Irurueta, D Ramasamy, M Datta, M Rangarajan · The Journal of Physical Chemistry C 120 (39), 22398-22406, 2016 |
| 33 | 2015 | Multiscale multiphysics model for hydrogen embrittlement in polycrystalline nickel S Jothi, TN Croft, SGR Brown · Journal of Alloys and Compounds 645, S500-S504, 2015 |
| 32 | 2014 | Finite element microstructural homogenization techniques and intergranular, intragranular microstructural effects on effective diffusion coefficient of heterogeneous … S Jothi, TN Croft, SGR Brown, EA de Souza Neto · Composite Structures 108, 555-564, 2014 |
| 30 | 2017 | Corrosion inhibition on mild steel by phosphonium salts in 1 M HNO3 aqueous medium N Palaniappan, LR Chowhan, S Jothi, IG Bosco, IS Cole · Surfaces and Interfaces 6, 237-246, 2017 |
| 30 | 2016 | Localized microstructural characterization of a dissimilar metal electron beam weld joint from an aerospace component S Jothi, T Sebald, HM Davies, ED Reese, SGR Brown · Materials & Design 90, 101-114, 2016 |
| 23 | 2015 | Coupled macroscale-microscale model for hydrogen embrittlement in polycrystalline materials S Jothi, TN Croft, SGR Brown · International journal of hydrogen energy 40 (6), 2882-2889, 2015 |
| 19 | 2023 | Grain boundary and triple junction characteristics analytics of additive manufactured Inconel 625 superalloy using selective laser melting C Pleass, S Jothi, M Krishnan · Materials Science and Engineering A 869, 144744, 2023 |
| 19 | 2019 | Neodymium-decorated graphene oxide as a corrosion barrier layer on Ti6Al4V alloy in acidic medium N Palaniappan, IS Cole, F Caballero-Briones, S Manickam, C Lal, S Jothi · RSC Advances 9 (DOI: 10.1039/c9ra00106a), 8537–8545, 2019 |
| 19 | 2015 | Meso-microstructural computational simulation of the hydrogen permeation test to calculate intergranular, grain boundary and effective diffusivities S Jothi, N Winzer, TN Croft, SGR Brown · Journal of Alloys and Compounds 645, S247-S251, 2015 |
| 18 | 2018 | Electrodeposition of Tin-Bismuth Alloys: Additives, Morphologies and Compositions AR Rajamani, S Jothi, M Datta, M Rangarajan · Journal of Electrochemical Society 165 (2), D50-D57, 2018 |
| 17 | 2019 | Highly branched PtCu nanodandelion with high activity for oxygen reduction reaction T Yang, D Pukazhselvan, EL da Silva, MC Santos, L Meng, D Ramasamy, S Jothi, V Graça, S Shi · International Journal of Hydrogen Energy 44 (1), 174-179, 2019 |
| 17 | 2015 | Modelling the influence of microstructural morphology and triple junctions on hydrogen transport in nanopolycrystalline nickel S Jothi, TN Croft, SGR Brown · Composites Part B: Engineering 75, 104-118, 2015 |
| 10 | 2021 | Artificial neural network modeling to evaluate and predict the mechanical strength of duplex stainless steel during casting T THANKACHAN, KS PRAKASH, S JOTHI · Sādhanā 46 (4), 197, 2021 |
| 3 | 2019 | Microstructural influence on damping for Friction stir welded Aluminum plates K Vijayan, S Jothi, A Pal · Procedia Structural Integrity 14, 642-648, 2019 |
| 3 | 2018 | Studies on the effect of weld defect on the fatigue behavior of welded structures R Murugan, PR Venugobal, TP Ramaswami, S Chinnusamy · China welding 27 (1), 53-59, 2018 |
| 1 | 2023 | pH‐Dependent Synthesis of Pure‐Phase BiI3 and BiOI by Electrodeposition** GM Kumar, DS Varma, S Jothi, D Ramasamy, JCC Abrantes, M Rangarajan · ChemistrySelect 8 (29), e202302055, 2023 |
| 1 | 2017 | Machine Learning for Degraded Mechanical Properties of Metallic Materials Due to the Presence of Hydrogen SJT Thankachan, KS Prakash, CD Pleases, D Rammasamy, ... · Int. J. Hydrogen Energy, 2017 |
Conference papers and presentations (17)
Newest first.
Reports, preprints and thesis (7)
Including the EU FP7 MultiHy deliverable and the PhD thesis (Swansea University). Newest first.
| Cites | Year | Title and source |
|---|---|---|
| – | 2023 | Metadata optimisation of RAG for STEM Chatbot S Jothi · 10.5281/zenodo.10447789, 2023 |
| – | 2023 | Retrieval Augmented Generation(RAG) for AEEM domain specific personalised Tasks S JOTHI, D Ramasamy, JM Olive · https://doi.org/10.5281/zenodo.10396326, 2023 |
| – | 2023 | Data Index Tuning for Retrieval Augmented Generation S Jothi · 10.5281/zenodo.10447764, 2023 |
| – | 2022 | Electrochemical Synthesis of Bismuth Iodide and Bismuth Oxyiodide MK Govindaraju, DS Varma, S Jothi, D Ramasamy, J Abrantes, M Rangarajan · ChemRxiv, 2022 |
| – | 2022 | Investigation of Grain boundary and triple junction characteristics of Inconel625 manufactured using selective laser melting S Jothi, C Pleass, M Krishnan · 10.5281/zenodo.7386523, 2022 |
| 6 | 2015 | Multiscale modelling and experimentation of hydrogen embrittlement in aerospace materials S Jothi · PQDT-Global, 2015 |
| 4 | 2015 | EU FP7 multi-scale modelling hydrogen embrittlement-MultiHy S Jothi, TN Croft, SGR Brown, E de Souza Neto · Swansea University deliverable 6, 2015 |
Industry and proprietary work (7)
Listed by title only: work for industry partners or kept private, so the details are not public.
| Year | Title and source |
|---|---|
| 2018 | Corrosion resistance behaviour of reduced graphene oxide /epoxy nanocomposite coating for Magnesium. SJ Kavimani.V, Soorya Prakash.K, Titus Thankachan, Gunashri, Sathish P, R ... |
| 2016 | Effect of homogenization on Mechanical Property of newly developed self-healing alloy S Jothi, T Thankachan |
| 2016 | 2D Graphene Nitrogen Based Compound Functionalized for Titanium Alloys Anticorrosion coating in 1M H2SO4 Medium S Jothi, R Chowhan, IG Bosco, IS Cole |
| 2016 | Electrochemical Sensing of Ferulic Acid using RGO-TiO2 Nanocomposites S Jothi, M Rangarajan |
| 2014 | Multiscale multiphysics atomistic-meso-continuum critical dislocation method for hydrogen embrittlement S Jothi |
| 2013 | ASYMPTOTIC HOMOGENIZATION MODELING OF MAGNETO-ELECTRIC SMART COMPOSITES DA Hadjiloizi, AL Kalamkarov, S Jothi, AV Georgiades |
| – | Audio Processing and Speech Recognition |