I design the object and then make sure it can actually be built.
I'm Krithick — a product designer and mechanical engineer in Ingolstadt, finishing an MSc at Technische Hochschule Ingolstadt. My background runs both ways: a mechanical engineering degree and BAJA competition fabrication on one side, industry design projects with Rohde & Schwarz and tesa on the other. That combination is the whole point. I can take a concept from a sketch through CMF and CAD to a structure that tolerates load, tooling and a service technician.
Three doors above, because the work genuinely splits three ways. Pick whichever matches the role you're hiring for.
Objects that earn their place on the counter.
Product development and concept design, from fluid-dynamics-led kitchen appliances to consumer audio. MSc at Technische Hochschule Ingolstadt.
Blenders had stopped being interesting to look at. Standard machines shear on one axis, leave fibrous fruit gritty, and hide the whole process behind opaque plastic. The brief was a premium countertop juicer that extracts better and is worth watching.
Three axes of counter-rotating titanium blades inside a reinforced polycarbonate sphere. The opposing rotation cancels torque — so the machine stays still — while generating high-shear turbulence that breaks down fibre other blenders leave behind.
No buttons. Rub the orb to start, adjust speed, or shut down — haptic control borrowed from a lamp, not an appliance. Interior LEDs light the vortex so the extraction reads as theatre.
Two blade sets, opposed.
Primary blades drive the bulk flow; secondary blades run counter on offset axes, folding the flow back into itself. The result is chaotic advection rather than a single circulating vortex — the same principle industrial mixers use to homogenise viscous media.
Everyone owns a dispenser. Nobody likes using one.
Research turned up the same five complaints in every workplace we looked at — and none of them are about adhesion. They are about the object.
Brand recognition, an iconic form, smart handling, premium materiality and creative expression inside the brand guidelines. Their own design characteristics were the scorecard we designed against.
The team split the characteristics into four design routes. I took Robust and drove it to a resolved product — Sporty Block — then built Flick Flack as the counter-proposal: what the same brief looks like if the dispenser moves instead of sits.
Three materials, three arguments.
Weight low, blade high, hand nowhere near it.
The body is a single wedge rising to the cutting edge, so the pull direction and the geometry agree: you draw tape up the slope and tear against a ridge that is already braced. The heavy base sits under the roll, where the moment is, which is what stops the slide that makes dispensers a two-hand job.
Everything brand-carrying is a recess, not a sticker: a chamfered arrow around the cutter, an embossed outline tracking the silhouette, and the logo dropped into a flat facet so it survives handling.
Flick Flack — a dispenser that folds shut.
Two arms on one hinge: open, the upper arm holds the roll and presents the cut end at hand height; closed, it collapses flat over its own blade and becomes a desk object with a pen channel. The knurled hinge disc is the grip, the brake and the brand mark in one part.
Printed at size to check the two things renders lie about: whether the slope reads as a pull direction without instruction, and whether the cutter sits where a thumb expects it. The profile got shallower and the cutter ridge moved forward after this one.
Presented to tesa alongside the team's other three routes as part of the final review.
That the fix for a dispenser is mass and geometry, not features. Nothing here moves, latches or adjusts — the improvement is entirely in where the weight sits and which way the surface points.
A weighted print in the real composite to test tip-over at the worst pull angle, a moulding review on the recessed detailing, and a hinge-life test on Flick Flack before either goes further.
Structure that has to hold, ship, and be serviced.
Load-bearing assemblies, mounts and enclosures — designed in Fusion 360 for real fabrication, tolerance and teardown. Industry work with Rohde & Schwarz, plus student competition structures.
Airframes built for the practice of building airframes.
Concept aircraft modelled in Fusion 360 and SolidWorks, rendered in KeyShot. No brief, no client — surfacing practice that turned into a set of designs I keep returning to.
LS-01 Dark Wolf
A sixth-generation fighter designed around one mission the others treat as an edge case: the super-low-altitude strike. The fuselage and intake system cut the radar cross-section hard while still feeding the engines the air they need down there.
Flying a strike profile at treetop height is a different design problem from flying one at altitude. The threat is short-range and it is looking up at you against ground clutter, so the payoff is in the planform seen from above and in never presenting a corner reflector.
So the whole aircraft became one continuous surface: a chined nose blending into a tailless delta, no vertical break anywhere along the top, and the intake moved to the shielded upper deck where the wing itself hides it from below.
ASF-01 Dracula
A multirole combat UAV. Its structure allows manoeuvre that a manned aircraft could not survive — take the pilot out and the airframe stops being designed around a human's tolerance, so the shape is free to go somewhere else entirely.
Almost every line on a fighter traces back to the person inside it: the canopy bulge, the seat rails, the g-limit that caps how hard you can turn. Delete the pilot and all of that goes with them.
What replaced it is a joined-wing layout — a forward lower plane running back into a swept upper plane, closing the structure into a loop. It is stiff for its weight, it carries control surfaces at both ends of a long moment arm, and it produces the one thing this aircraft is for: a rate of change of direction that would put a crew out cold.
Sly Fox
A low-observable rotorcraft whose job is to deliver precision munitions at low altitude and high speed — then leave. Armament is a co-pilot-operated 30 mm autocannon and four concealed missile pylons.
Attack helicopters are usually drawn as gun platforms with a rotor bolted on: slab sides, external stores, every surface a return. This one starts from the fuselage instead — a single continuous body with the tail boom flowing out of the spine rather than being attached to it.
The weapons live inside that body. Pylons sit flush until they are needed, the cannon is chin-mounted and slaved to the gunner in the front seat, and the stub wings do lift work rather than acting as a rack. Speed and a small return come from the same decision.
F/A-18 Legacy Hornet
Somebody else's aircraft, rebuilt from reference surface by surface. No livery, no markings, no camouflage — the point was never the picture. It was whether I could make the surfaces meet cleanly enough to survive a mirror.
Paint hides everything. A camouflage scheme will happily cover a surface that kinks at a blend or a fillet that changes radius halfway along, and you will never know. So this one stayed bare and got rendered on a near-mirror finish instead.
The stripes running down the wing and along the spine are the reflected horizon. Where they stay straight and evenly spaced the surface is continuous; where they would kink or bunch, the model is wrong. It is a zebra-stripe check done in the renderer rather than in the CAD viewport — the most honest way I know to publish a surfacing exercise.
Specific detail on the QPW500 platform — RF specifications, board layouts, dimensions and internal documentation — is covered by a non-disclosure agreement with Rohde & Schwarz and is not published here. What is shown is my own structural design work and the capability it demonstrates: geometry is generalised and no proprietary electronics data appears. I'm happy to walk through the full detail in an interview under NDA.
The team studied the scanner from both sides of the glass: attendees who slow down because nothing tells them what to do, and operators watching a queue build while they wait on results. Alongside those experience problems sat two blunt physical ones — the machine is difficult to install and difficult to move.
The installation and transport half. If the structure inside the housing were modular rather than monolithic, everything downstream gets easier: a service tech reaches one board instead of stripping a wall, a crew crates the machine instead of freighting it whole, and the scan geometry still holds when it is stood back up.
Five parts, five separate jobs.
Boards in front, everything else behind.
Each front-end board sits on standoffs behind its own mount plate, so it is located by the plate rather than by the housing. The cut-outs in the plate clear the connectors and leave a service hand a way in without pulling neighbouring boards.
The rest of the electronics goes on the back face of the column. Front side is the array; back side is the plumbing. That split is what makes a maintenance visit a one-panel job.
The tower next to the machine became the floor of the machine.
Four GPUs, a main board and their cooling live in a drawer in the base plate. Air comes in the front through three fans and leaves through the vented flank; the interface plate puts every connector on one edge, so the machine has one face to plug into and one drawer to pull.
A scanner that travels as parts, not as freight.
Presented to the Rohde & Schwarz team in May 2025 as part of the project's final review, alongside the UX and post-control work from the rest of the team. The endoskeleton carried the transport and maintenance argument: a machine that can be crated, carried, re-erected and serviced without the housing doing structural work.
FEA on the column joints under the crowd-protection case, a fastener count reduction pass on the head plate, and a tolerance stack from base datum to the top board mount to prove the calibration claim rather than argue it.
One race, two fluids, one transform point.
Designing the vehicle meant designing the sport first. The aircraft take off from the water surface and fly an aerial course that rewards agility over top speed; at the halfway marker they switch propulsion and wing geometry for submerged travel, then run a set of underwater checkpoints past seabed obstacles to the finish. The canyon leg is a straight nod to Top Gun.
A delta because both fluids want the same thing.
The body is a single tapered volume with a low-drag nose and no separate fuselage-to-wing joint — the same section that cuts air cleanly is the one that has to cut water. A thin swept delta gives the wing area to fly the canyon leg and folds into a small frontal profile for the submerged run; the single fin keeps yaw authority in both.
The canopy is a long closed bubble, blended flush, because on this vehicle any step in the surface is a cavitation problem as well as a drag one.
One hybrid engine, two modes.
The proposal is a hybrid jet that transitions from air-breathing to water propulsion on the same shaft, running on hydrogen fuel cells — so the transform point is a change of intake path rather than a change of vehicle.
Same airframe, opposite personalities.
The whole proposal on one sheet.
The board carries the race map, the phase sequence, the propulsion diagram and the two liveries — the deliverable was a sport someone could be sold on, not just a vehicle.
That a vehicle concept is only as good as the world it races in. The constraint that made the design interesting — one body, two fluids — came from writing the format first.
Design the transform sequence properly — what physically moves, in what order, and how the intake seals. Right now it's asserted on the board and not resolved in the CAD.
Two closed volumes, joined by as little as possible.
Endurance prototypes are shaped by what the rules leave alone. Zeke takes that to its conclusion: the wheels sit inside two enormous enclosed pods, the cockpit is a narrow spine dropped between them, and the whole middle of the car is a tunnel rather than a body. There is no grille, no lamp cluster, no visible intake — the front is a single blade.
The proportion only makes sense from above.
Seen flat, the car reads as four covered wheels and a blade. The tricolour brushstroke was drawn to work in this view first — it crosses the door and rear pod at the same angle the surfaces break, so the livery describes the shape instead of decorating it.
White body, a hand-drawn saffron-and-green stroke, and a race number where a sponsor would sit. It was made for the reveal render rather than for a season — the point was a car that looks like it belongs to somewhere.
Resolve the cockpit ingress — right now the canopy is one surface with no answer for how a driver gets in — and take the underbody from a shape to a plausible floor.
Everyone redesigns these for adults.
A rugged Bluetooth barrel is already a good product for a child: it rolls, it survives being dropped, it has no screen and no account. What stops it being a kid's speaker is entirely surface — black fabric, grey rubber, a logo in chrome. So I left the architecture alone and rewrote the skin.
A print that survives being looked at every day.
That a new audience doesn't always need new engineering. The change here is entirely CMF and graphics on an existing platform — the cheapest kind of product decision, and the one most often skipped.
Put it in front of actual children — print sizing and glow brightness are the two things I'd expect to get wrong — and check the loops against a real strap and a real neck.
A bookend that travels with the collection.
A small shelf holds three books one week and nine the next, and every fix for that is ugly: a loose bookend that falls over, or a fixed divider that wastes the rest of the board. Here the end panel runs in a channel cut into the base — push it up against the spines and the books are clamped; slide it back and the shelf is open again.
The figurine is the handle. It leans into the panel with its whole body, which is both how you grip it and why the object is worth having on a desk: the mechanism is the story.
That adjustability doesn't have to look technical. One channel and one character do the job that catalogues solve with brackets.
Build it in wood. The open question is friction — the panel needs to hold a row of books without being stiff enough that a child gives up, and that's a test, not a render.
{{ npTitle }}
{{ npSummary }}
{{ npB1Body }}
{{ npB2Body }}
{{ npB3Body }}
One bevel pair doing three jobs.
Where the tail gearbox sits in the drive line, and what happens inside it once the shaft arrives.
Specific detail on the RUAV-200 tail gearbox — gear data, tolerances, load cases and the test parameters agreed with the Rotary Wing R&D Centre — is covered by a non-disclosure agreement with Hindustan Aeronautics Limited and is not published here. What is shown is my own rig design work and the capability it demonstrates: the specimen is represented generically and no proprietary gearbox data appears. I'm happy to walk through the full detail in an interview under NDA.
The tail gearbox sits at the top of the tail pylon and drives the rotary rudder. It does three jobs at once: cuts shaft speed by roughly 3:1, turns the drive through 90°, and passes pitch changes from the flight control system out to the tail rotor blades. It is splash-lubricated and air-cooled — no pump, no radiator, nothing to add weight at the far end of the tail.
That combination is why it earns its own rig. A gearbox changing both speed and direction under a live control input has more ways to fail than a bench can guess at, so the rig had to reproduce the real load path rather than simply spin the input shaft.
Airworthiness is a measure of how fit an aircraft is to fly, and every rotorcraft component is rated for a set number of flight hours and cycles. An unchecked mechanical fault does not stay mechanical for long — it becomes a reliability problem and then a safety one.
So the crucial components are run on the ground before they are installed. The brief was to design that ground apparatus: validate the quality of the tail rotor gearbox, learn what its maintenance actually demands, and design and analyse a functional rig for the specimen.
Six parts between the wall socket and the specimen.
Each one exists to make the test survivable: step the speed up, hold the shaft steady, absorb what vibrates, and break before the gearbox does.
Everything on one welded table.
Motor and driver pulley up top, driven pulley and bearings on the cross member, then the fuse, the coupling and the specimen in a straight line. Sandwich-type dampers under all four feet.
The FEM optimisation of the loaded components and the final load calculations were the next stage of the plan of action, alongside the submission to RWRDC. Those results are not shown on this page due to the company's NDA.
The mechanical fuse is the part I would take further. It protects the specimen, which is the expensive thing in the room — but a shear pin sized for one torque ceiling makes the rig single-purpose. A replaceable, calibrated fuse cartridge would let one rig cover a family of gearboxes.
An all-terrain vehicle built from nothing by a student team, then driven for five days over rock, sand and a water crossing. I was lead designer. Sixth overall out of fifty teams, eighth in design and analysis.
FMAE Baja is run by the Fraternity of Mechanical and Automotive Engineers, and Season 3 finished at their Moto Park outside Hyderabad in February 2020. The scoring is split: you are judged on the engineering design and analysis you can defend in front of a panel, and then on whether the thing you built survives days of rough terrain.
Those two halves pull against each other constantly. Every gram you take out of the vehicle to make it quicker is a gram of material you have to argue for on paper — and then trust when the car lands off a jump with a driver inside it.
Being lead designer on a student team is less about drawing and more about arbitration. The roll cage, the suspension pick-up points, the driver's space and the fire-extinguisher mount all want the same volume, and every subsystem lead is certain theirs should win. My job was deciding, in order, and keeping the decisions consistent enough that the fabricators could keep cutting tube.
It is the most honest design education I have had. Nothing hides. If a bracket is wrong you find out in front of twenty people with a grinder in their hands.
Everything you designed, being disagreed with.
Water crossing, loose sand, and a trail narrow enough that the front wheels do the steering whether you ask them to or not.
Twenty-odd people, one vehicle, and a deadline that does not move. Worth putting on the page because none of the above was done alone.
Digital Tools
Software I build so the hardware work goes faster. One project so far, and it is the large one: an environment that takes an invention from first thought to the point where it can be defended.
The product's own screens, rebuilt here to click through. Sample data throughout — nothing live.
The inventions Desk-E manages are real and private, so every screen on this site runs on illustrative sample data — an electric-aircraft-propulsion scenario authored for exactly this purpose. No real invention titles, scan domains or patent findings appear here, and nothing on these pages is live.
Five stages, each one a gate.
Every agent returns proceed, hold or stop. A hold is not a failure — it is the system declining to guess.
Agents never call each other.
All cross-agent data moves through one shared dossier — the Invention Record. Each agent reads it and appends to it, and that is the only channel.
It sounds like a small decision and it is the one that makes the thing survivable. A direct call between agents means a failure in one becomes a failure in all of them; a shared record means a broken agent produces a gap you can see rather than a cascade you cannot.
A consistent refusal to let the system fabricate.
The feature list is not the interesting part. These five decisions are.
Tools split in two. In some the model proposes and the user edits — PESTEL, SWOT, pros and cons. In the rest the user supplies the numbers and the agent may only challenge them: cost–utility, the Pfeiffer portfolio, TRL, payback. An agent inventing your cost figures would be fabrication, not assistance.
CARAT was first implemented with axes I had invented. When the real Department of Energy framework surfaced, the whole tool was rebuilt around the published seventeen-dimension rubric and its scoring table. Keeping the plausible-looking version would have been cheaper and dishonest.
You are allowed to disagree with a computed score — drag the bubble. But the override is badged with what the assessment actually computed, the bubble gets a marker, and one click restores it. Disagreement is a stated position, not a quiet edit.
The radar hover popup exists partly as proof of provenance: headline, detection date, source, why it matters. A segment with nothing real to show stays empty rather than being filled with something plausible.
FOR-E ships with a disclaimer modal and a badge that does not go away, because continuous live monitoring is not running yet. A tool that overstates its own certainty is worse than one that does less.
An AI decision tool must make the provenance of every figure visible at the point of use.
Click into the instrument.
Rebuilt to explore, not to work. Editing affordances are shown and do nothing.
Is now the right moment?
A radar, used as a radar. Angle is the category, distance is time to impact, colour is priority, and a contact's nose points where it is heading. Hover one to see where it came from.
A list cannot show you that something is closing.
Foresight output is usually a feed: twelve headlines, each with a severity tag. The problem is that a feed flattens the two things an inventor actually needs — how soon, and getting worse or better. Both are spatial, so the display should be spatial.
Borrowing the whole grammar of an aircraft radar — heading, velocity line, closing versus receding, an alarm ring for the thing that needs attention now — means the screen is legible at a glance to anyone who has seen one, and rewards a longer look without needing a legend.
Provenance at the point of use.
Every contact carries its detection date, its source and a plain-language reason it is on the face at all. That is not a nicety — it is the condition of the tool being trustworthy. A mark on a radar is an assertion, and an assertion you cannot trace is a guess with better graphics.
The corollary is the rule that a segment with nothing real to show stays empty. An evenly-populated radar would look more finished and would be a lie about how much is known.
Is the idea worth pursuing?
Nine tools in three gated phases. Each column is its own carousel — step through the tools with the arrows or the dots. Every tool renders a real chart and shows the arithmetic underneath it.
Two kinds of tool, and the difference is not cosmetic.
In PESTEL, SWOT and the argument balance the model drafts and you edit — it is faster to correct a wrong first pass than to face an empty grid. Each of those tools says so in its own header.
In cost–utility, the Pfeiffer portfolio, TRL and payback, you supply the numbers and the agent may only challenge them. An agent inventing your cost figures is not assistance, it is fabrication with a progress bar.
A score with no arithmetic is an opinion.
Every tool prints the calculation under the chart — the inputs, the operation and the threshold. It costs three lines of space and it is the difference between a number you can argue with and a number you have to trust.
It also catches the near-misses. The strengths and weaknesses tool here nets to exactly zero, and the payback crossing falls between two months. Both get reported as what they are rather than rounded into a verdict.
Can it be protected?
Six scored dimensions, a headline verdict, and seven sections that say what the score is made of. The interesting part is what it refuses to round: a strong novelty score sitting next to a middling freedom-to-operate risk is not one number.
The shape is the finding.
Averaging these six dimensions gives 68.7, which sounds like a pass and tells you nothing. The useful reading is the notch: novelty and market breadth are strong, freedom to operate is not, and those two facts imply completely different next actions.
A polygon makes an uneven profile impossible to miss in a way a single score or a stack of bars does not. The mean is still printed, deliberately underneath, because someone will ask for it.
It is not a legal opinion, and it says so.
Every section names its evidence: how many documents, which classification, whether a claim is granted or merely published. The freedom-to-operate section will not return "clear" — the best it offers is a risk band with the reason attached.
That restraint is the point. A tool that tells an inventor their idea is free to practise, on the strength of a search, is doing something worse than nothing.
{{ cur.summary }}
{{ para.t }}