Website:
bhatiyaniai.com
Job details:
About the RoleWe are hiring a 3D Reconstruction & Generative AI Engineer to build an end-to-end system that can generate high-precision 3D jewellery models from images and text prompts.
This role goes beyond classical CV — you will design systems that combine:
- 3D reconstruction (multi-view / single-view)
- Generative AI (text-to-3D / image-to-3D)
- RAG pipelines for design retrieval and conditioning
The final output must be manufacturing-grade meshes suitable for gold casting / 3D printing.
Key Responsibilities3D Reconstruction & Geometry- Build pipelines for:
- Multi-view → 3D mesh reconstruction
- Single-view → 3D with learned priors
- Implement:
- Point cloud generation, fusion, and meshing
- Surface reconstruction and refinement
- Ensure:
- High geometric accuracy (sub-mm level)
- Clean topology (watertight meshes)
Generative AI (Core Focus)- Develop and fine-tune models for:
- Text-to-3D generation
- Image-to-3D generation
- Work with:
- Diffusion models (2D → 3D lifting)
- NeRF / Gaussian Splatting-based methods
- Latent 3D representations
- Integrate:
- Style control.
- Conditional generation (user constraints)
RAG (Retrieval-Augmented Generation) Systems- Build RAG pipelines for:
- Retrieving existing designs / CAD assets
- Conditioning generative models with reference designs
- Design:
- Embedding pipelines (image + text embeddings)
- Vector databases for design search
- Enable:
- “Generate similar to this design” workflows
- Hybrid pipelines:
- Retrieval → conditioning → generation → refinement
- Optimize for:
- Low-latency retrieval
- High semantic relevance
Model Optimization & Deployment- Optimize models for:
- GPU efficiency (multi-GPU setups)
- Batch inference pipelines
- Deploy:
- APIs for generation (image/text → 3D)
- Scalable backend systems
Post-Processing & CAD Readiness- Implement:
- Mesh cleanup (hole filling, smoothing)
- Topology correction
- Conversion to CAD-friendly formats (STL/STEP)
- Ensure outputs are:
- Print-ready
- Structurally valid for casting
Required SkillsCore- Strong experience in Python + PyTorch
- Solid understanding of:
- 3D geometry (meshes, point clouds, SDFs)
- Multi-view geometry & camera calibration
- Rendering and differentiable rendering
- Experience with:
- PyTorch3D / Open3D / Trimesh
- 3D reconstruction pipelines (SfM, MVS, NeRF)
Generative AI- Hands-on experience with:
- Diffusion models (Stable Diffusion, latent diffusion)
- GANs / 3D generative models
- Understanding of:
- Text-to-image / text-to-3D pipelines
- Conditioning and prompt engineering
RAG & LLM Systems- Experience building:
- RAG pipelines (retrieval + generation)
- Familiarity with:
- Vector databases (e.g., FAISS, Milvus)
- Embedding models (CLIP, multimodal embeddings)
- Ability to:
- Combine structured (CAD) + unstructured (images/text) data
- Design semantic search systems
Preferred / Bonus Skills- Experience in:
- Jewellery design / CAD workflows
- Blender / MeshLab / Rhino / Fusion 360
- Familiarity with:
- Vision-language models
- Grounding models (e.g., object-aware generation)
- Experience with:
- Multi-GPU training (important for your setup)
- Diffusion + 3D hybrid pipelines
Expected Outcomes- Build a system capable of:
- Text/Image → high-quality 3D jewellery
- Retrieval-assisted generation (RAG-powered design system)
- Deliver:
- Production-ready meshes
- Consistent design quality across variations
Candidate Profile- 2–6+ years in:
- Computer Vision / Generative AI / 3D Graphics
- Strong in both:
- Research understanding
- Production deployment
Nice-to-Have Project Experience- Text-to-3D systems
- RAG-based generative pipelines
- High-detail reconstruction (jewellery / mechanical parts)
- CAD-integrated AI systems
Compensation- Competitive (based on experience and depth in GenAI + 3D)
- BETWEEN 4-8 LPA, BASED ON PERFORMANCE.
Click on Apply to know more.