LILA·SUTRAलीला सूत्र
THE COMPLETE PORTFOLIOA journey through timeNewest first, from today's generative-AI systems back to where it started. Videos link out to Drive for playback; image frames are being curated in.
← BACK TO PROFILE
2024-25Generative AI & World-Building

Sr AI Generalist at ADSMN, plus late-stage AjnaLens R&D. Client systems up top; personal craft and research below.

CLIENT WORK · ADSMN INTERACTIVE
MaazaFILM · FACE-PRESERVED AVATARS

Four films for Maaza's summer campaign - real base plates with AI-swapped brand avatars, identity locked frame-to-frame, on a multi-GPU diffusion + video pipeline. Each set below: plate, avatar and finished film together.

Multi-GPUFace-preserveVideo / CV
SET 01PLATE → AVATAR → FILM
▶ PLAY
BASE PLATE
AI AVATAR
SET 02PLATE → AVATAR → FILM
▶ PLAY
BASE PLATE
AI AVATAR
SET 03PLATE → AVATAR → FILM
▶ PLAY
BASE PLATE
AI AVATAR
SET 04PLATE → AVATAR → FILM
▶ PLAY
BASE PLATE
AI AVATAR
FevikwikCONVERSATIONAL AGENT · BRAND VOICE

KwikGPT - an agentic LLM that became the brand's voice: fast, cheeky, on-brand replies at scale, taken all the way to three aired films.

Agentic LLMBrand voicePrompt-sys
▶ PLAY
Aired film 01
▶ PLAY
Aired film 02
▶ PLAY
Aired film 03
KWIKGPT OUTPUTS
VadilalVOICE CLONE · DIALOGUE

'Zara Muskuraiye' - character-driven spots with voice-clone consistency and clean dialogue. Final samples landing soon.

Voice cloneDialogue
FINAL SAMPLES LANDING SOON
De BeersJEWELLERY LoRA · PRODUCT + ON-MODEL

A jewellery LoRA and placement pipeline for INDRA, De Beers' AI platform: a single product input is restyled across creative backdrops, upscaled and outpainted, then composed onto models, identity of the piece preserved throughout.

LoRAStyle transferOutpaintOn-model
INPUT → STYLE OUTPUTS → ON MODEL
INPUTSTYLE · NATURESTYLE · 3DON MODELON MODELNecklaceRingEarringBraceletNose pin
Milka × PokoraAI MUSIC FILM · VOICE + LIP-SYNC

A music-led film with French artist M. Pokora: generated lyrics and a cloned vocal (ACE-Step), mastered, then lip-synced to picture. International campaign.

Music AIVoice cloneLip-sync
FOREIGN CAMPAIGN
▶ PLAY
Sample 01
▶ PLAY
Sample 02
Yippie · Better WorldKIDS' SUPERHERO AVATARS · SAFE OUTPUTS

Kids become their own superheroes - face-preserved avatars in two art styles for boys and girls, with pose edits (Qwen), background removal and vision-agent safety checks across the 'Better World' campaign.

Face-preservePose-editSafetyVision
BETTER WORLD · ENDED
BOYS · STYLE I
BOYS · STYLE II
GIRLS · STYLE I
GIRLS · STYLE II
2022–24AjnaLens - where 3D became AI

Two years at AjnaLens (DimensionNXG). I arrived for 3D and left an AI generalist - here strongest-first: the systems, then the XR products I storyboarded and shipped, then the personal generative craft underneath it all.

▶ PLAY
Draw3DSketch-to-3D: rough a shape by hand, get a usable model back.
▶ PLAY
AjnaKriti · World GeneratorText-driven 3D world generation - the R&D seed that became BuilderVerse.
▶ PLAY
AjnaSwastik · Show and TellDiffusion prototypes - LoRA, ControlNet, IP-Adapter, ComfyUI.
▶ PLAY
AjnaAvatar · MetaHuman SDKMetaHuman avatars in Unreal, wired to spoken audio and an LLM.

I trained a FLUX LoRA on photographs of the AjnaXR Pro headset, then ran inference across styles and themes, turning a single product into an endlessly re-stylable asset. Every frame below is a generated output of the trained model.

A full VR training simulation for ITI fitter apprentices on the AjnaXR headset, storyboarded end-to-end - onboarding and profiles, the Foundation and Industrial learning paths, tool selection, machine inspection, and hands-on tool use.

01 · On open
02 · After intro AV
03 · After login
04 · P1 user & app info
05 · P2 foundation path
06 · P3 industrial path
07 · U1 account
08 · U2 replay intro
09 · U3 credits & info
10 · F1 tools & domains
11 · F2 interactive UI
12 · F3 enter workshop
13 · B1/B2 foundation buttons
14 · S1 wearing the suit
15 · Tool selection
16 · Machine inspection
17 · Using the tool
18 · Detailed tool usage
FOUNDATION MODULES · BY TOOL FAMILY

The home and launcher for AjnaXR's learning platform, shown at CES 2023, storyboarded from a first pass to the final flow.

V2 · FINAL
01
02
03
04
05
06
07
08
09
10
11
V1 · FIRST PASS
01
02
03
04
05
06
07
08
09

Model-agnostic by habit - a few favourites from FLUX.

A self-initiated worldbuilding series - उरात्म Uratma, the “mighty souls”: enigmatic cosmic beings at the crossroads of the mortal and the divine. First explored on Stable Diffusion 1.5, then re-forged on SDXL as Divergent Speciation.

SD 1.5 · THE ORIGINAL SERIES
SDXL · DIVERGENT SPECIATION

The deepest, oldest archive - 3,000+ frames from 2022–23. A curated cut of the strongest themed sets; the full library is on Drive.

2020–23VR simulations & real-time

Training sims, interaction design and capstones across PupilFirst, E3 and MAGES, on Quest 2 and mobile XR.

▶ PLAY
SanskritVRQUEST 2 · LIP-SYNC
▶ PLAY
Inventory Management VRWAREHOUSE INTERACTION
▶ PLAY
Deep Dive RescueVR CAPSTONE
Industrial VROPEN FOLDER ↗Custom Unity ProjectOPEN FOLDER ↗
2019–22Foundations - 3D & animation

Where it began at MAGES Singapore: modelling, animation exercises, and the first showreel.

▶ PLAY
3D Animation ShowreelSHOWREEL
▶ PLAY
Bouncing BallsEXERCISE 01
▶ PLAY
Damped PendulumEXERCISE 02
▶ PLAY
Obstacle CourseEXERCISE 03
▶ PLAY
Vanilla Walk CycleEXERCISE 04
▶ PLAY
Run CycleEXERCISE 05
▶ PLAY
Smooth Criminal MoonwalkANIMATION
▶ PLAY
Sci-Fi Gun LoopANIMATION
▶ PLAY
InvasionFILM · 2020
▶ PLAY
No-Good Do-GooderFILM
Model Sketch & ShadingIMAGE
NEBULA