Research & R&D · Zazhree AI Engine
ZazimFind's R&D programme combines computer vision, neural rendering, mobile AR, and cloud inference to solve a commercial problem that smaller businesses face every day: how to present products with enough realism to earn trust and drive conversion.
R&D journey
The programme has progressed through data collection, model development, mobile engineering, and structured testing. Each phase narrows the gap between research feasibility and commercial utility.
The R&D programme began with server setup, dataset assembly, and initial neural models for segmentation and 3D structure inference. The first working pipeline produced textured meshes from front-facing photographs on a desktop setup.
The team built an Android application that captures product images, sends them to a cloud-hosted AI engine, and renders the resulting 3D model with rotate-and-zoom interaction. ZazimFind Ltd was incorporated in March 2026.
A 100-user pilot confirmed the interface was intuitive and useful. Local retailers integrated product images for live testing, surfacing practical limits in cluttered backgrounds, uneven lighting, and reflective surfaces.
The next stage is a closed beta with 10–15 retail partners, improving reconstruction accuracy, UI clarity, and system reliability before public launch.
The production release will focus on GDPR compliance, better inference orchestration, annual billing, and the first public customers.
The roadmap includes multi-room walkthroughs, improved scene reconstruction, and an iOS release using a cross-platform mobile stack.
Technical architecture
The Zazhree engine is a cloud-backed pipeline with a lightweight mobile client. Compute-heavy reconstruction scales independently in the cloud while the user experience stays simple.
Real-time radiance field rendering for fast, photorealistic results.
Zazhree uses explicit 3D Gaussian primitives where appropriate so that rendering can remain fast, interactive, and mobile-friendly. This is especially valuable when the final output must be streamed efficiently to smartphones and web viewers.
A diffusion-guided optimisation loop for image-to-3D reconstruction.
The pipeline uses pretrained diffusion priors to guide 3D generation from sparse or single-image inputs. This avoids the need for paired 3D training data across every product category and helps the system generalise to the long tail of SME inventory.
Foreground isolation and structure inference before geometry generation.
The first stage segments the object from the background, estimates a coarse depth map, and infers material cues such as roughness and reflectance. This structured representation improves downstream geometry quality.
Physically based rendering preserved across glTF and USDZ.
Zazhree prioritises materials that matter to buying decisions: fabric weave, metallic reflection, and realistic scale. Assets are exported in standard formats so they can be consumed by Android, iOS, and web-based experiences.
Validation metrics
The current prototype is already producing measurable outputs. These metrics determine whether the system is ready for broader release, scaling, or further research investment.
Technology Readiness Level
TRL 4
Component validation in a laboratory environment.
Training images
12,000+
Across 50 product categories.
Mean Chamfer Distance
0.34
Held-out test set result.
PSNR
28 dB
Reconstructed output quality.
Generation time
< 5s
Targeted inference time after optimisation.
Key technical learnings
Foreground segmentation must happen before 3D reconstruction, because cluttered retail scenes confuse geometry prediction. Synthetic lighting variation and background substitution improve robustness. Inference time was reduced by tightening the optimisation loop and using faster rasterisation, while edge-preserving losses improved the visual confidence of the final model.
Market research
ZazimFind is built around a documented market need. These findings explain why the product is timely, not just technically interesting.
House of Commons Library
5.7M
UK SMEs, accounting for 99.9% of all private sector businesses and a substantial share of employment and turnover.
UK Business Data Survey 2024
32%
Of UK businesses had no website at all; the digital adoption gap remains material among smaller firms.
Rithum 2025 Returns Report
41%
Reduction in return rates reported when moving from static images to interactive 3D with AR try-on.
BVDW Whitepaper 2026
94%
Conversion lift documented for 3D-enabled product pages relative to static imagery.
Ofcom Media Nations 2024
93%
UK adults who own a smartphone, confirming a mature mobile environment for AR commerce.
MarketsandMarkets
40%
Forecast CAGR for the global AR retail market, supporting timing for the category.
Commercial interpretation
UK smartphone penetration, growing mobile commerce, and fast-rising AR retail adoption make 2026 a practical launch window. The product is aimed at SMEs that need conversion gains and return reduction without enterprise-level overhead.
Policy alignment
The platform aligns with UK government priorities around SME digitisation, AI adoption, and dual-use innovation. This matters for programme fit, credibility, and the availability of non-dilutive support.
DSIT / UK Government
Positions AI as a primary lever for broad-based economic growth and explicitly supports the UK becoming an AI maker rather than an AI taker.
SME Digital Adoption Taskforce
Recommends closing the gap between awareness and application, with high-impact digital tools aimed directly at small firms.
MOD / DASA
The defence research environment supports technologies that work in both civilian and military contexts, including AR-based simulation.
Risk and mitigation
The research programme is designed with clear mitigations for adoption, accuracy, funding, compliance, and talent risks.
Small businesses often cite cost, complexity, and vendor trust as the main barriers to digital adoption.
Mitigation
Low entry pricing, a free trial, direct onboarding, and a simple capture workflow reduce the burden of adoption and make the product easy to try without technical staff.
Generative 3D systems can struggle with reflective surfaces, clutter, and poor lighting.
Mitigation
A segmentation stage, augmentation strategy, early user feedback, and staged validation keep quality improvement tied to real retail conditions.
Seed and angel markets are selective. Early-stage capital can be difficult to secure without milestone discipline.
Mitigation
Non-dilutive paths through Innovate UK and DASA are part of the roadmap, alongside a milestone-based hiring plan and a tightly scoped seed ask.
AI, data protection, and invention disclosure all require disciplined process.
Mitigation
Trade secrets protect the technical moat, while GDPR, security controls, and a formal disclosure workflow reduce legal and operational risk.
Five-year roadmap
The roadmap connects the R&D programme directly to commercial milestones: launch, scale, enterprise expansion, and strategic dual-use applications.
2026
Validate
MVP refinement
Closed beta with SMEs
Seed funding close
GDPR and security hardening
2027
Launch
Public release
iOS version
Shopify / WooCommerce integrations
Analytics dashboard v1
2028
Scale
2,000 SME customers
Multi-angle capture improvements
API access
International pilot markets
2029
Expand
Property and interiors enhancements
Film pre-vis tooling
Defense sandbox trials
Enterprise licensing
2030
Consolidate
Profitability
Wider UK adoption
Strategic partnerships
Long-term platform moat
Next step
This page connects naturally with your existing header, head, and footer. It supports investor, grant, and technical evaluation use cases without requiring a separate theme.
