Building LLM systems that survive audit, evals, and real users.
Over the last four years I’ve shipped LLM systems in finance, e-commerce, and regulated compliance. I build directly against provider SDKs rather than through frameworks, because the parts that matter (prompts, data, evals, latency) are the parts frameworks don’t help with.
Right now I’m a KTP Associate at Pimberly × University of Essex, building an EU compliance platform that produces Digital Product Passports.
Before that, I led the AI team at Elixir that shipped a RAG financial chatbot on WhatsApp, and migrated a negotiation chatbot at Nibble from rule-based NLU to LLM-driven generation. My first IEEE paper, on MicroVM honeypot isolation, is out at CSR 2026.
On the side, I take on a small number of consultancy engagements: audits, prototypes, and full builds for teams who want production-ready AI without the framework sprawl. Happy to talk through specifics on a short intro call.
Sole engineer on an end-to-end compliance platform covering the full lifecycle: VLM-OCR ingestion of regulatory PDFs, structured extraction with field-level confidence, a reviewer-facing audit trail, and Digital Product Passport generation for the EU ESPR. Also designing the evaluation harness for a comparative study of open-source LLMs on compliance extraction.
Firecracker-based honeypot isolation with per-attacker VMs. 242 ms median startup, 106 MB memory footprint, zero overhead for legitimate users, validated at 100 concurrent sessions. First IEEE publication.
Senior-most developer leading the in-house team that shipped an AI-driven personal finance platform. Designed the intent + entity pipeline, built RAG over financial regs and product data, and set up daily pipelines, observability (Loki / Grafana) and automated conversation testing.
Migrated an e-commerce negotiation chatbot from rule-based NLU to LLM-driven generation in production, versioning models against each other. Implemented an academic negotiation strategy for live pricing, and designed a generic agent layer over the existing NLU.
Most AI features stall between ‘working demo’ and ‘in production.’ I’ve worked on that gap in finance, e-commerce, and regulated compliance, and I can help you do it faster.
A short, hands-on review of an existing or planned AI feature. I look at the data, the pipeline, the evals, and the production risks, and come back with a prioritised list of what to fix and what to build.
End-to-end: scoping, data ingestion, retrieval or agent pipeline, eval harness, deployment on your infrastructure. Delivered by me, documented well enough that your team can own it after handover.
After a build ships, I can stay on for a few hours a week: monitoring, regression gates, model upgrades, and tuning retrieval as the corpus grows. For systems you depend on but don't want to babysit.
Half-day and full-day sessions for engineering teams: practical AI engineering, eval design, retrieval, and agent loops. Hands-on, no slide-heavy theatre. Remote or on-site in the UK.
Every engagement starts with a free 30-minute call to see whether it’s a fit. Pricing is agreed per engagement and shared on the intro call.
Shariq consistently demonstrated outstanding technical expertise, strong problem-solving capabilities, and an unwavering dedication to excellence. He made significant contributions to the development and optimisation of our AI stack.
Shariq consistently displayed great abilities and attitude. Smart, friendly, always looking to learn, and supports his fellow developers with good advice. Very knowledgeable and capable with generative LLMs. A pleasure to work with, highly recommended.
A reliable and valuable team member. Shariq's positive attitude, strong loyalty, and excellent technical skills in AI are invaluable. His technical expertise contributes significantly to the success of our product.
Looking to hire, consult, or trade ideas? Email is the fastest route, or book a short call directly.