Deployed a production GenAl extraction pipeline on AWS (Lambda, S3, SNS, IAM, CloudFormation, CloudWatch) that validates aircraft task cards across several document types, replacing manual line by line review. Worked directly with BMD business users to translate evolving business rules into extraction logic and prompt revisions, and validated output against their judgement. Built the manhour billing matcher for non-routine task (NRT) work on A350 MRO, benchmarking several information retrieval methods and LLM based pipelines, including an agentic variant, then tuning retrieval weights and confidence thresholds with BMD users against real edge cases. Meaningfully lifted matching accuracy and reduced billing leakage at low cost per card. Worked on prompt engineering for a RAG based quality incident querying tool, building a ground truth evaluation set and diagnosing several failure classes across structured evaluation rounds, then resolving them through a multi-step querying redesign. Prepared a fine-tuning pipeline for a BGE embedding model for aviation domain retrieval, generating labelled training triplets through co-occurrence and clustering analysis for triplet loss training with hard negative mining, to improve retrieval on domain vocabulary that off-the-shelf embeddings handled poorly. Consolidated a large aircraft maintenance rate card down to a fraction of its original size using TF-IDF with clustering and LLM assisted terminology standardisation, cutting lookup cost and latency while keeping the rate logic auditable. Ran a cost and performance benchmarking study across candidate LLMs for production workloads, comparing accuracy, latency, and cost so model selection became an evidence based decision rather than a default. Won third place at the SIAEC Group ITD Hackathon with PartPilot, an agentic parts and tooling extraction tool, and built a Telegram based RAG quiz bot on Claude and ChromaDB at Anthropic's Push2Prod hackathon. Authored the company-wide Prompt Library and GenAl Capability List, both adopted organisation wide, created instructional videos and led knowledge-sharing sessions for senior leadership.

Advised early-stage biotech and medtech founders on operations, market positioning, and funding strategy, using structured research and quantitative analysis to replace assumption-led decisions with evidence clients could defend in front of investors and partners Ran in-depth research across technology, process optimization, market, and financial domains, translating dense scientific and technical material into business terms non-specialist stakeholders could act on Produced tailored reports and presentations integrating scientific data with commercial insight, helping clients refine business models, strengthen go-to-market plans, and attract the partners and investors they were targeting Worked in a cross-functional team to evaluate operational scalability and funding options, coordinating analysis from members with very different technical backgrounds into one coherent recommendation