Technical portfolio spanning production applications, BI middleware, custom AI pipelines, and practitioner-led engineering publications on arXiv & SSRN.
Production SaaS products, client BI architectures, autonomous agent pipelines, and research data engineering infrastructure.
Architected custom Node.js middleware bridging Toggl, Absence.io, and internal CRM tool APIs into a centralized Power BI analytics engine for executive leadership.
Architected NVision's enterprise Salesforce Sales Cloud, Data Cloud, and Tableau BI analytics infrastructure from scratch across Europe & North America, establishing a single source of truth for master data with 1-click access to relevant commercial insights, custom automation flows, PandaDoc quote integrations, Apex/Lightning components, and automated ETL pipelines.
Production self-hosted n8n workflow automation platform orchestrating multi-service webhooks, scheduled cron triggers, and custom API integrations across local containerized nodes.
Self-hosted Munich rental intelligence system with Custom Tuned AI Agents monitoring the rental market every 3–7 minutes, scoring ads via multi-stage LLM chains, and pushing instant Web/Mobile/Telegram app alerts.
Cloud-hosted OpenAI-compatible LLM gateway and custom 15 KB static chat SPA on ARM64 cloud architecture, featuring Caddy reverse proxy, PocketBase forward_auth, and streaming SSE responses.
Data ingestion infrastructure built for a TRR 266 working paper, extracting machine-readable corporate annual report sections for NLP analysis.
Multi-pipeline agentic system that ingests job descriptions, extracts structured experience evidence, scores ATS keyword coverage, and generates tailored DOCX/LATEX CV drafts with real-time SSE progress streaming.
An open-source agentic pipeline that turns raw meeting transcripts into validated Jira epics, user stories, acceptance criteria, and technical task trees via Atlassian REST API.
Self-improving custom tuned AI Agents monitoring 60+ platforms including job boards and enterprise ATS platforms, scoring role match via MPNet bi-encoders and Thompson-sampling feedback loops.
Google Drive → Notion byte-integrity migrator designed for unattended cloud deployment on lightweight VM infrastructure, executing 6-pass deterministic triage, property-based testing, and chunked uploads.
Autonomous daily pipeline capturing GitHub commit diffs and transcripts, synthesizing citable evidence manifests and draft posts into Notion with three-layer alerting.
Automated iPhone backup pipeline storing 3 independent copies of photos, videos, and files across immich (dedicated VM), self-hosted cloud drive with 5-tier verification gates.
Eight independent Scriptable widgets that turn abstract 'time remaining' into a daily visual signal. Each one hand-rolls its own date math and gradient rendering with zero dependencies, and every widget below runs as a live port in the browser.
Practitioner-led empirical studies and open publications on arXiv and SSRN, synthesizing production learnings across software testing, BDD optimization, LLM verification, and AI energy economics.
Co-authored systematic literature review (PRISMA 2020) of 54 studies on LLM-based test oracles, published in IEEE Access, introducing a source-of-authority taxonomy. Organises the field by where oracle verdicts derive their authority, a dimension missing from prior secondary studies.
arXiv paper reporting on a production LLM-integrated, multi-market rental-search assistant whose 1,553-test automated suite passed continuously yet shipped user-facing defects. Introduces a four-seam defect classification framework based on analysis of 252 bug-fix commits.
Evidence-anchored synthesis paper reconciling AI's dual energy narrative: while efficiency gains reliably cut energy and carbon intensity per task, cheaper compute triggers Jevons rebound effects that expand absolute energy demand.
Co-authored arXiv paper mining BDD subscenario refactoring candidates across 5.3M slices using ML classifiers benchmarked against LLM judges. XGBoost classifier (F₁ = 0.891) beat a tuned rule baseline and two open-weight LLMs (GPT and Ling) at p < 10⁻⁴.
Co-authored arXiv paper presenting an automated HTTP API quality assessment framework. Playwright-based browser instrumentation captured 108 HAR files across 18 production websites, applying 8 heuristic anti-pattern detectors to produce a composite quality score per site.
Ask Bilal’s AI Clone