Working Style & Expectations
We are looking for committed, highly motivated professionals who take ownership seriously. This role requires consistent collaboration during U.S. Eastern Time business hours, reliability in scheduling, and strong follow-through. Being on time, prepared, and diligent is essential — this is a senior role on a system that underpins real operations, not an experimental project.
We value people who are excited by hard problems, communicate clearly, and treat commitments as non-negotiable.
1. AI/ML Developer Model Training & Fine Tuning
2. Data Model & Architect
3. Developers Microservices / API ETL & Integration developement
We are building an end-to-end system that turns unknown physical parts into known, sellable products — automatically. The system ingests parts through vision and sensors, resolves identity using AI and structured knowledge, and publishes those parts into a fully integrated commerce, logistics, and search ecosystem.
This is not a research role. It is a production systems role spanning computer vision, large-scale data ingestion, retrieval systems, cloud infrastructure, and e-commerce integration.
What you will build
- A learning ingestion pipeline that identifies unknown physical parts using images, 3D data, and metadata, and improves continuously through human confirmation
- Vision and retrieval systems using multimodal embeddings and vector search to match parts across large catalogs
- A RAG-based knowledge system that ingests, parses, and normalizes manufacturer PDFs, schematics, and parts manuals into APIs
- Event-driven services that publish identified parts into inventory, commerce, shipping, and ERP workflows
- Search-optimized part representations engineered to be discoverable by Google, analytics systems, and downstream AI models
Why this is hard — and interesting
- Parts are visually similar, worn, incomplete, and inconsistent across manufacturers
- Manufacturer data is fragmented, unstructured, and often only available as PDFs and diagrams
- The system must operate across perception, inference, search, commerce, and logistics — not just ML
- Identification must be probabilistic, explainable, and correctable, while downstream systems require deterministic truth
- The platform must learn continuously without breaking accounting, inventory, or customer trust
This is a real-world intelligence problem, not a model-training exercise.
Technical environment
- Google Cloud Platform (Cloud Run, Compute Engine, GPU instances, Cloud Storage)
- MongoDB Atlas for operational and catalog data
- Vector search and retrieval systems for embeddings and similarity matching
- Event-driven and API-first architecture
- Computer vision, multimodal embeddings, and RAG pipelines in production
Required experience
- Senior-level experience building production AI or data-driven systems end-to-end
- Strong background in computer vision, retrieval systems, or multimodal ML
- Experience designing cloud-native systems on GCP or similar platforms
- Deep Python expertise; comfort owning architecture and trade-offs
- Ability to connect ML outputs to real business systems (commerce, ERP, logistics)
What success looks like
- Unknown parts can be ingested, identified, and published without manual data entry
- Manufacturer documents become structured, queryable knowledge
- Identified parts flow cleanly into e-commerce, search, and fulfillment systems
- The system improves over time without increasing operational risk
