founders-factory - AI software Engineer - Project Tricorder
Requirements
• We don't expect one person to be expert in everything below. Strong candidates will be deep in one workstream and credible in the other. • Strong Python and data engineering skills, with a track record of shipping production data or ML pipelines, not just research notebooks. • Hands-on experience building ML evaluation frameworks, benchmarks or test harnesses, ideally for vision, video or language models. • Experience running data labelling and training-data preparation, including annotation tooling and quality control. • Working knowledge of databases and data modelling. • Practical NLP experience: entity extraction, entity linking or document parsing. • Comfort handling sensitive data under data protection and clinical safety constraints. • Pragmatism and speed: you can work with ambiguity, make good build-vs-buy calls and ship in weeks, not quarters. • Experience with SNOMED CT, ICD-10, dm+d or other NHS clinical terminologies and coding. • A background in health-tech, clinical informatics or another regulated domain such as defence. • Experience with video data at scale (frame sampling, temporal annotation, multimodal datasets). • Familiarity with FHIR or HL7 interoperability standards. • Experience working with LLMs and VLMs, including edge or on-device deployment. • Early-stage startup experience, especially as a founding or early engineer. • PROBABLY NOT THE RIGHT FIT IF • You need detailed specs and a large team around you to be effective. • You prefer pure research over building tools people use every day. • This is a high urgency opening, and we are flexible to consider applicants either on a more flexible consultancy agreement (rolling by mutual agreement, with a competitive day rate); or a long term permanent employment contract with shares as part of the package and scope to grow into a longer-term role as Tricorder builds its engineering team.
Responsibilities
• You'll own two workstreams, working day to day with the founder and the FF build team. • 1. DATA PIPELINE AND EVALUATION HARNESS • Build the data foundations that tell us whether our models are good enough for clinical use. • Data labelling: design labelling schemas and guidelines, set up tooling, and run labelling with clinical input, including quality checks and inter-annotator agreement. • Data processing: build reproducible pipelines to ingest, clean, de-identify and transform multimodal data (video, sensor streams, text). • Video data organisation: structure, version and catalogue video datasets so clips, annotations and metadata stay searchable and traceable. • Training dataset preparation: assemble balanced, well-documented train, validation and test splits, with clear lineage from raw data to model input. • Eval harness: build an automated harness that benchmarks vision-language and NLP models against clinically meaningful metrics, tracks regressions and supports build-vs-buy model decisions. • 2. CLINICAL KNOWLEDGE BASE • Build the structured clinical knowledge base that turns unstructured input into coded, interoperable records. • Database design: design the schema and stand up the data store for clinical concepts and their relationships. • Clinical terminologies: ingest and map TC3, SNOMED CT, ICD-10 and related NHS standards, including cross-mappings between them. • Natural language processing: extract entities and relationships from clinical text and link them to coded concepts. • Document handling: parse and structure clinical documents (PDFs, notes, forms) into the database. • Clinical knowledge: work with clinicians to make sure the database reflects real-world clinical reasoning and pre-hospital workflows. • By the end of the first 3 months, we'd expect a working eval harness in use for model decisions, a labelled and versioned core dataset, and a first version of the knowledge base linked to the product.
Apply in one click
Upload My Resume
Drop here or click to browse · Tap to choose · PDF, DOCX, DOC, RTF, TXT