A. J. R. Lewis
AI Architect
Hi, I'm Alex.
I architect and implement LLM-powered systems, workflow engines, and internal tools tuned for enterprise constraints.
From extra-galactic astrophysics to enterprise AI delivery, I bring scientific rigor, engineering depth, and fast execution.
In front of ALMA whilst observing with LABOCA, Atacama Desert, Chile, 2016.
My Background
I moved from academia into tangible software engineering, then into what I care most about: architecting practical AI systems.
Astrophysicist
Built large-scale data workflows and statistical models in high-volume research environments.
Download PhD ThesisSoftware Engineer
Shipped production forecasting systems supporting 1M+ SKU/store combinations with adaptive retraining.
AI Architect
Designed and implemented LLM-powered workflows and automation, now leading enterprise-ready AI system design, integration, and delivery strategy.
What I Offer
I use AI to augment teams, reduce repetitive work, and improve delivery quality. I am in favor of scaling output without default headcount growth, but not reducing existing teams in favor of automation.
Consulting
I assess workflows and constraints to identify where AI can improve ROI, then work with your CTO to define and implement the right delivery path.
Map workflows, constraints, and baseline metrics before committing to build.
Rank use cases by ROI, risk, and implementation complexity.
Define architecture, rollout phases, and ownership with your CTO and technical leads.
You get a clear, measurable AI roadmap your team can execute with confidence.
Implementing
I build high-performance, provider-agnostic systems designed for total ownership. By avoiding proprietary silos and integrating automated evals from day one, your infrastructure stays portable and scalable.
Production-grade FastAPI, Next.js, and PostgreSQL (pgvector).
Docker-containerized by default for cloud or on-prem deployment.
Custom drag-and-drop workflows for fast iteration without architectural drift.
You own the code, the data, and the keys, with a production-ready system instead of a black-box subscription.
Teaching
I mentor your team with practical handover, documentation, and training so they can operate and extend what we build with confidence.
Runbooks, architecture notes, and operating checklists tailored to your team.
Hands-on sessions for developers and operators across workflows, evals, and day-to-day usage.
Internal ownership from week one, with clear support boundaries and maintenance rhythms.
Your team can run, improve, and extend the system with AI as a productivity tool, not a replacement for people.
Selected Projects
View more on GitHubFinance
Gresham House
Engaged as a consultant to design and implement maintainable AI architecture for long-term handover, starting with stakeholder workflow discovery and operating constraints. Built a Python/TypeScript foundation that coordinates MCP interactions across Claude, custom web, and API applications, linking Salesforce and Snowflake for faster, data-grounded RFP support.
Target outcome: sustainable internal ownership, faster RFP turnaround, and more reliable data-backed responses.
Impact Signals
- Integrated 2 core platforms: Salesforce and Snowflake.
- Coordinated MCP workflows across Claude, web, and API channels.
- Architecture designed for internal handover and long-term ownership.
Downing
Downing's digital team needed to triage Microsoft Teams requests and route each digital issue to the right person. I delivered an Azure-native LangGraph orchestration pilot with Teams HITL routing, secure state management, and provider-flexible automation to test whether handoff and response times could be improved.
Pilot showed faster routing and clearer ownership of incoming requests.
Impact Signals
- Azure-native LangGraph orchestration pilot delivered.
- HITL routing through Microsoft Teams implemented.
- Pilot validated faster handoff and clearer request ownership.
Agency / Ops
Digital Speed
Teams needed reliable operational answers in Slack; built a TypeScript/Next.js AI agent with handbook and tool integrations for fast, in-context support, plus a GitHub-driven company-as-code ingestion flow that re-chunked changed docs and refreshed embeddings for RAG.
Faster answers, fewer context switches.
Impact Signals
- Operational assistant embedded directly inside Slack.
- Company-as-code ingestion keeps knowledge base synchronized.
- RAG context refresh reduces stale or outdated responses.
Startups
Hyperhumans
Lead discovery was manual and noisy; built the Neuralead FastAPI backend for automated enrichment, scoring, and monetized API delivery at scale.
Higher-quality leads and more relevant conversations.
Impact Signals
- Lead enrichment and scoring pipeline automated end-to-end.
- Monetized lead intelligence exposed through API delivery.
- Better lead relevance for outbound conversations.
W4lkies
Open SourceA grassroots local project and one of my favourites: built W4lkies, a full-stack platform unifying bookings, customer and dog profiles, invoicing, and expenses, with continuous UI and workflow improvements to support day-to-day operations.
Lower admin overhead, smoother daily operations, and ongoing usability gains.
Impact Signals
- 4 core workflows unified: bookings, profiles, invoicing, and expenses.
- Operational admin overhead reduced for a local business.
- Continuous UX improvements shipped against live usage feedback.
References & Case Studies
Detailed case-study metrics and client references are available during qualified conversations. Some outcome data is shared under NDA.
Request referencesGet in Touch
Let's scope your next AI system.
Share your constraints, timelines, and goals, and we'll map an architecture and delivery plan your team can execute.