Aerospace engineer. I build engineering tasks, reference solutions and graders that test AI models on real design work.
Bengaluru, India · MSc Aerospace Engineering, University of Salford · LinkedIn · dineshnatarajan02@gmail.com · CV (PDF)
- Engineering tasks for AI training and evaluation. A task is a problem statement in the voice of a stress or manufacturing office, a disclosed engineering standard, a verified reference solution, and an automated grader that accepts any valid answer and rejects the textbook shortcut. I build all four, then calibrate difficulty against repeated agent runs and review the grader for ways a submission could score without doing the engineering.
- Track record. Six Hardware/CAD tasks accepted on Handshake AI's Project Insight, a Terminal-Bench 3 benchmark, in October 2026. Each ran in a separate verifier container with held-back cases and exact geometric grading. The last two were not solved in any of five trials by the benchmark's frontier model. Earlier, CFD and structural reinforcement-learning tasks at Alignerr (Labelbox), and aerospace solution review at Innodata.
- The engineering underneath. Finite-element analysis, CFD and aerodynamics, flight dynamics, fatigue and damage tolerance, B-rep geometry and GD&T.
| Project | What it is | Result | Stack |
|---|---|---|---|
| ReconLoop · live demo | Settlement reconciliation agent graded against an answer key on held-out batches; a model breaks ties only where rules cannot | 83 tests, CI fails on a single wrong link across 20 held-out batches | Python, Gradio, Pyodide |
| Aerodynamic surrogate | Neural-network surrogate for airfoil Cl, Cd and L/D, benchmarked against three other regressors on a dataset calibrated to the XFLR5 study | 612 points, L/D R² 0.998, MAPE 2.9 % held out, 2.7 to 6.7 % on unseen Reynolds numbers, about 10⁶× faster than the solver | Python, scikit-learn |
| Engine bracket: FEA, topology optimisation, GD&T | Aircraft engine mount bracket in SAE 4130 from validated analysis to a toleranced, inspectable definition | 22 % lighter at a factor of safety of 1.73; worst-case ±0.35 mm and RSS ±0.206 mm stack-ups | ANSYS Mechanical, CATIA V5, ASME Y14.5-2018 |
| Natural-laminar-flow airfoil study | NACA 63-215 and 64-210 against NACA 2412 across Re 500k to 3M, with 3D wing and stability derivatives | About 44 % higher L/D and 42 % lower drag at cruise | XFLR5, lifting line, vortex lattice |
| Cessna 172 longitudinal flight dynamics | 4-state small-perturbation state-space model with eigenvalue analysis and four elevator inputs | Short-period ωn 5.60 rad/s, ζ 0.58, inside MIL-F-8785C; 6 of 6 V&V checks | MATLAB, Simulink |
| Twin-boom surveillance UAV | Conceptual design to a 3D-printed model in a three-person MSc team | 2.2 m span, 2.8 kg take-off weight against a sub-5 kg brief, 0.5 kg payload | SolidWorks, XFLR5 |
| When | Role |
|---|---|
| Sep to Oct 2026 | Task author, Hardware/CAD, Handshake AI, Project Insight. Six accepted terminal-based tasks with separate verifier images, held-back cases pinned by SHA-256, pytest graders, and a sandbox that runs the tool under test as an unprivileged user. |
| Sep 2026 to present | Aerospace engineering expert, AI training, Innodata (via Upwork). Review and rate aerospace question, solution and answer sets, with a written justification for every rating. |
| Jun to Sep 2026 | Reinforcement-learning task author, engineering simulation, Alignerr (Labelbox). Wing-planform and OpenFOAM tasks behind solver-health gates, OpenSeesPy retrofits, deterministic graders with hidden evaluation cases, difficulty calibrated against frontier-agent runs. |
| May 2023 to Sep 2025 | Procurement and inventory operations supervisor, Tops, Manchester. Led a 21-member operations team. |
| Oct 2020 to Apr 2021 | Product design engineer (intern), Black Space Research and Development. Shell-element FE model of a rear underrun protection device in HyperMesh; three regulatory load cases with contact in Abaqus/Standard; governing case 234.6 MPa against 230 MPa yield. |
- MSc Aerospace Engineering, University of Salford, UK, 2023 (2:1). Dissertation: structural analysis of an aircraft engine bracket by the finite element method.
- BTech Aerospace Engineering, Karunya Institute of Technology and Sciences, India, 2021 (2:1).
- Master Diploma in Product Design, CADD Centre, 2022 (certificate) · Airbus Supply Chain Specialist, Airbus Beyond, 2026 (certificate) · Machine Learning Specialization, DeepLearning.AI and Stanford Online, 2026.
Python (NumPy, SciPy, pandas, scikit-learn, pytest) · Bash and Linux · Docker · Git and GitHub Actions · ANSYS Mechanical · Abaqus/Standard · Altair HyperMesh · OpenFOAM · OpenSeesPy · XFLR5 (XFoil) · AeroSandbox · CadQuery and OCP (OpenCascade) · PTC Creo · CATIA V5 · MATLAB/Simulink
Open to contract and full-time work in engineering simulation, CAD and AI evaluation. Email is the fastest way to reach me.