MSc Applied Artificial Intelligence & Data Analytics (Merit) · University of Bradford
Bradford, UK · open to relocation
Portfolio Case studies Jobzyl, live LinkedIn Email ORCID Springer publication
I build retrieval systems that pair vector embeddings with knowledge graphs, benchmark predictive models on large real-world datasets, and take both to production: retrieval design, backend APIs, deployment.
- 🔬 Most recent research: FinLaw-UK, a graph-augmented RAG system over UK financial regulation where every citation is resolved against a Neo4j knowledge graph before it reaches the user, and the system refuses rather than answering when retrieval is weak.
- 🧾 Re-measured my own evaluation pipeline post-submission, found two reported metrics were regex shape-checks rather than correctness measures, and published the correction alongside the true graph-verified rate. The measurement-integrity section is above the fold in the repo, not in an appendix.
- ⚙️ Shipped an AI voice-agent platform that handled 2,100+ outbound calls and cut mean call latency 54% (2.4s → 1.1s).
- 📄 First-author and corresponding-author on a peer-reviewed Springer publication (ICSMAI 2024, Morocco).
- 🧪 Benchmarked 11 classifiers over 253,680 CDC BRFSS records for diabetes risk, with resampling confined to the training folds.
- 🎯 Currently open to AI/ML engineering roles and funded doctoral positions in the UK and EU.
🔎 Jobzyl · jobzyl.com
Live job-search aggregator serving a 2M+ posting index. Queries 20+ boards (Indeed, Adzuna, Reed) with parallel fan-out on miss, per-source timeout isolation, and Server-Sent Events (SSE) streaming for sub-1.5s search results. Features an advanced ATS-style CV matcher and semantic vector ranking engine to score role-fit. Enforces zero-trust privacy via in-browser parsing (Fernet encrypted), alongside server-side deduplication, Kanban-style application tracking, and automated email digest alerts.
TypeScript Next.js React Python FastAPI PostgreSQL Supabase AWS
🏛️ FinLaw-UK
Graph-augmented RAG over UK financial regulation
Hybrid BM25 + BGE-small retrieval with reciprocal rank fusion, 2-hop Neo4j traversal, and Mistral 7B-Instruct served locally via Ollama. Every citation is resolved against the graph before the answer ships; provisions absent from it are flagged rather than passed through.
Python Neo4j Ollama FastAPI Flask Docker
0.76 RAGAS faithfulness, 0.74 answer relevance, and 0.68 legal completeness — the figures stable across both evaluation tracks and both scoring implementations.
110-item harness, 10 curated items · MSc dissertation
🎙️ VoiceFlow
Secure Speech Transcription Pipeline
Engineered a secure call-export pipeline for Retell AI outbound voice agents, ensuring zero-trust data privacy by transcribing sensitive customer recordings locally via an open-source Whisper large-v3 model.
Python FastAPI PyTorch Whisper Next.js
Multi-threaded async fetching · Live SSE progress
Processed 2,100+ production calls
11-classifier benchmark + screening tool
Diabetes risk on CDC BRFSS 2015, comparing random over-sampling, SMOTE and ADASYN with resampling confined to the training folds. Shipped as a lab-free, 19-question screening app.
scikit-learn pandas Flask React
Random Forest best on ROC-AUC and sensitivity
253,680 records · 14% positive class
Flask + React
The model behind my ICSMAI 2024 paper
Four regression models compared, winner served behind a Flask form.
Random Forest R2 0.8569 (MSE 0.0027)
First-author, corresponding author
Full case studies, with architecture diagrams and the decisions behind them → hammadahmad.co.uk/projects
Ahmad, H. (first & corresponding author), Khan, M.U., Azam, M. (2024).
Comparative Analysis of Machine Learning Methods for Enhancing Sleep Efficiency and Prediction.
In: Serrhini, M., Ghoumid, K. (eds) Advances in Smart Medical, IoT & Artificial Intelligence, ICSMAI 2024.
Information Systems Engineering and Management, vol 12, pp. 3-15. Springer, Cham.
Presented at ICSMAI 2024, Saidia, Morocco, 18-20 April 2024.
DOI: 10.1007/978-3-031-66854-8_1
Software & data release. The FinLaw-UK implementation and its evaluation harness are open under MIT at github.com/1oNN/finlaw-uk: a 110-item UK financial-regulation QA structure spanning factual questions, document tasks and case scenarios, of which 10 items are fully curated with gold answers and required citations and are the meaningful evaluation set. The remaining rows are template stubs and are documented as such in the repo's known limitations.
Machine learning & data
XGBoost LightGBM pandas NumPy MLflow Hugging Face
Retrieval & knowledge
RAG Hybrid retrieval Cross-encoder RAGAS FAISS Neo4j Ollama
Engineering
Data & infrastructure
Founder & Sole Engineer · Jobzyl (self-employed) · Jun 2026 - Present · Remote
Engineered and operate jobzyl.com, a live job-aggregation platform serving a 2M+ posting index across 20+ boards.
- Cache-first reads with live parallel fan-out on miss, per-source timeout isolation, and SSE streaming for sub-1.5s retrieval
- Built an advanced ATS-style CV matcher and semantic vector ranking engine to score role-fit
- Enforced zero-trust privacy via in-browser parsing (Fernet encrypted), alongside server-side deduplication and Kanban tracking
- Next.js, TypeScript, React, Python, FastAPI, PostgreSQL, Supabase, AWS
AI / Machine Learning Engineer · Outlyst · Oct 2025 - Mar 2026 · Leeds, UK / Remote
Fixed-term contract. Built and deployed an AI voice-agent system for outbound calling (Retell AI, FastAPI): backend dialogue-flow logic, automated call flows, lead-qualification rules, gatekeeper detection and callback scheduling.
- Handled 2,100+ calls and cut manual calling workload by roughly 50%
- Profiled async I/O and connection pooling to reduce mean call latency 54% (2.4s → 1.1s)
- Built an internal micro-CRM for lead tracking, removing external CRM licensing costs
Research Assistant | Graph-Augmented LLM Engineering · University of Bradford · Jan 2025 - Sep 2025
Designed and evaluated FinLaw-UK: Mistral 7B served locally via Ollama, paired with a Neo4j knowledge graph, over the FCA Handbook, PRA Rulebook, FRC standards and statutory sources.
- Engineered the retrieval pipeline: clause-level segmentation, Sentence Transformer embeddings, BM25 + dense fusion by reciprocal rank fusion, cross-encoder re-ranking, and graph-grounded citation verification
- Built the evaluation harness, extending RAGAS with a custom legal-completeness metric that reproduces at 0.68 across both evaluation tracks
- Re-measured the submitted evaluation post-hoc, established that two reported metrics were format checks rather than correctness measures, and published the correction with the code
- Supervised by Dr Tillal Eldabi and Dr Irfan Mehmood
Research Intern, Data Science · COMSATS University Islamabad · Jul 2023 - Jul 2024
Benchmarked 11 classifiers for diabetes risk on 253,680 CDC BRFSS records, comparing random over-sampling against SMOTE and ADASYN for the 86/14 class imbalance, with resampling confined to the training folds.
- Analysed 20+ demographic, lifestyle and clinical indicators: age, general health, BMI, blood pressure and income emerged as the strongest correlates
- Deployed the winning model behind a REST API with SHAP-based interpretability
- Led the first-author comparative study of sleep-efficiency prediction published at ICSMAI 2024
MSc, Applied Artificial Intelligence & Data Analytics (Merit) · University of Bradford · 2024-2025
Dissertation: FinLaw-UK: A Graph-Augmented Retrieval Chatbot for Reliable and Transparent UK Financial Regulation
Modules included Artificial Intelligence and Data Science (79), Business Data Analytics (79), and Responsible AI: Ethics, Law and Governance (75).
BS, Bioinformatics · COMSATS University Islamabad · 2020-2024
Thesis: AI-Assisted Analysis and Prediction of At-Risk Diabetic Individuals, graded A.
Research interests
graph-augmented retrieval · LLM faithfulness evaluation · evaluation methodology for RAG · interpretable clinical modelling
Languages
English (IELTS 7.0, CEFR C1) · Urdu (native) · German (A1.2)
Right to work
UK Graduate visa to December 2027. Eligible to work now, no sponsorship required.
If you're working on retrieval, evaluation, or anything where a model needs to show its sources, I'd like to hear about it.