Consulting, AI implementation and automation. Co-founder of Navire. I audit business processes and build the automation that fixes them, including in regulated, data-sensitive environments where every threshold has to survive an audit.
Based in APAC · Malaysia, UTC+8 · Open to Singapore · French native, fluent English
I start by finding where the hours actually go, then build the system that removes the manual step. This is the bulk of the work and where the 10+ hours a week come back.
Operational audit · process automation · n8n and serverless · SOP enforcement
Agents that do a job rather than demo one: shared state across channels, retrieval that returns the right passage, and tooling the model can actually call.
Agentic workflows · LLM integration · RAG and vector search · MCP servers · voice AI and outbound calling
Connecting the systems a business already runs on, under real data protection constraints, and putting a usable screen in front of the result. Where the output has to be defended to a regulator, the thresholds and the failure modes get documented alongside it.
ERP / CRM / billing integration · GDPR-safe AI pipelines · AML / KYC screening and threshold calibration · beneficial ownership and PSC data · dashboards and admin panels · data governance and analytics
Generated cinematic film for technical audiences, written, cut and subtitled end to end. Published work has run to six figures of views.
Narrative shorts · brand films · vertical cuts · burned-in subtitling
Every engagement runs the same way: audit, scope, build, hand over. You own the system at the end, with no dependency on me.
An anonymisation pipeline that swaps names, SIRET numbers and IBANs for encrypted aliases before anything reaches the model, plus a module that turns raw Excel or CSV accounting exports into compliant FEC files. Demo above.
Sole technical lead on the client onboarding pipeline at Waverley Private Asset Partners, a multi-family office holding UHNW families through trust and corporate structures. Documents in, screened entities out, every threshold written down for external audit. Runs on the firm's hardware, so no client document leaves the perimeter.
The situation. Onboarding a single family meant reading trust deeds, articles, share registers, foreign registry extracts and source-of-wealth evidence by hand, then screening whoever that reading turned up. Standard investment files took two committee cycles to clear. No vendor tool covers the part that actually matters: working out who the screened parties are once a structure sits between the family and the asset.
What was built. A document pipeline that returns structured fields with a confidence score, each routed to a compliance reviewer, because KYC requires a person to sign off every field. Retrieval happens at page level before extraction rather than pushing whole documents at the model, which is where most of the accuracy came from. The same index answers questions over the deed corpus in natural language and cites the source clause. On top of it, each family's holding structure is reconstructed as a graph — settlor, trustee, protector, beneficiaries, underlying companies, Companies House PSC data — and resolved into one screened entity set. The PSC register is treated as the unverified self-reported source it is: corroboration rules, documented failure modes, and no direct-owner entry accepted as a beneficial owner on its own.
Calibration, not guesswork. Match thresholds on the firm's vendor sanctions and PEP feed were set by above-the-line / below-the-line testing against its own logged history: roughly 3,000 client and counterparty checks spanning two years, replayed and scored against the incumbent process. The recall condition was fixed up front — every standing PEP and sanctions match on the client book preserved, and no alert previously escalated by compliance allowed to auto-clear. Independently validated before go-live.
Where automation stops. Agreed in the same forum as the thresholds and documented for internal compliance sign-off and external audit: suspicious activity reporting kept out of scope entirely; beneficiary distributions, tax and succession excluded; low-confidence extractions and rejections go to mandatory human review and are never auto-declined. Record-keeping meets MLR 2017 and HMRC Trust Registration Service duties.
The result. Alerts requiring manual review down by about a third. Median file-open-to-decision time roughly halved since go-live, with standard investment files clearing in one committee cycle instead of two. The pipeline was triaged into the compliance workflow already in use rather than run alongside it, with runbooks written and the internal team trained; it now drives the annual review of the existing book as well as new files.
One Claude-powered agent behind voice, website chat, Telegram and WhatsApp: shared lead records, shared RAG knowledge base, shared scoring, plus fully automated outbound calling. Built for a B2B sales agency.
The situation. Four channels handled by four disconnected tools. The same lead could call and message on Telegram without anyone connecting the two. Outbound still meant a human dialing through a spreadsheet.
What was built. Thin channel adapters feeding one shared agent. Outbound campaigns dial through Retell AI and adjust lead scores from post-call sentiment. A 53 KB embeddable widget drops onto any website with one script tag.
The result. 4 channels unified in one pipeline, outbound qualification with no SDR involvement, live in partner testing across 7 Docker services with multi-tenant access control.
A serverless n8n system that audits three Google Calendars against strict meeting rules every 5 minutes and posts deduplicated alerts to Slack. Built for a US investment fund.
The situation. Protected mornings, internal-only and external-only days, buffers, daily caps. Enforcement was a tedious daily human review, often late, easy to miss.
What was built. A 5-minute polling audit engine with a Supabase-backed dedup layer, a daily 8:30 AM sweep with AI rescheduling suggestions, and a Friday preview of the coming week.
The result. The operations team stopped reviewing calendars entirely. Violations reach Slack within 5 minutes and duplicate alerts went to zero.
An MCP server that gives Claude long-term semantic memory: vector embeddings in Supabase pgvector, retrieved by meaning rather than keyword. Running in production daily.
The situation. AI coding assistants are stateless. Past bugs, decisions and conventions vanish when the session ends, and static notes files cannot be searched by meaning.
What was built. 8 MCP tools with token-capped recall, soft-delete memory expiry, and SQL-level filtering in a single round trip.
The result. 125 unit tests, Docker-deployed, and used in production every single day.
A phone agent that answers calls and takes reservations end to end. The dashboard replays every call with a transcript synced to the audio, flags interruptions, handles corrections mid-call, and produces an AI-written summary of each conversation. Built for a restaurant client. Demo above, English call.
A native macOS app that pairs a live system dashboard — RAM, disk, caches, processes — with a conversational agent. It explains what each file is and whether it is safe to remove, then cleans up on request, moving files to the Trash rather than deleting them outright. Demo above.
Site for a UK luxury bridal hair and makeup business. Live at thebridalproject.co.uk.
Showcase and shop for a French artisan crafting custom wargaming tables. Live at poildegrim.com.
Real-time indexer for native assets on Bitcoin L1. Live at universalscreener.xyz.
Cinematic short films for the Bitcoin developer niche, generated end to end and published on X, where the strongest have run between 80,000 and 140,000 views.
A 45-second brand film for Citadel: a marble senate, a dispute running across the ecosystem, closing on the client's mark. Built to hold a feed audience all the way to the logo.
140k views on X →A vertical short set in a neon Ethereum city, weighing the old governance era against what replaced it. Subtitles are burned in and keyed to the voice track, so it reads with the sound off.
Watch on X →Three minutes of narrative: a lone scavenger crossing a burning world to reach the Simplicity obelisk, and the ledger underneath it. Every shot, character and voice generated, then cut and subtitled.
London multi-family office, retained around five days a month alongside Navire. Sole technical lead on client onboarding and sanctions screening, working with the firm's compliance function, the client onboarding team and the investment committee.
Remote-first practice for process automation and pragmatic AI. Audits of client operations, then tailor-made automation across finance, HR onboarding, logistics and CRM, giving leadership teams back 10+ hours a week.
Identity verification platform in London. Ran production initiatives from scoping to release across engineering and machine learning teams, and contributed to Fraud Lab work on deepfake detection from 2023.
A 12-month data governance engagement in banking, then six months in energy framing an analytics roadmap. Data mappings, governance frameworks and business cases, plus workshops with client Data Officers and IT.
Twelve-month gap-year placement in the same practice, before joining full-time in 2020.
Grande École programme, specialisation in Business Analytics, with an exchange semester at the University of Michigan, Ross School of Business.
"I worked with Alexis for several weeks on a complex project involving the setup and integration of a CRM with our website, along with other important integrations for our business. He treated us with patience throughout the entire process, and the result is exactly what we wanted. I highly recommend him."
"I interviewed five different software developers for a solution to automate my processes. Alexis provided a simple solution that no one else considered. Very professional and very committed to providing a successful outcome."
"Professional and competent as always. Responsive to last-minute inquiries and revisions throughout the year. A reliable partner and a pleasure to work with."
Tell me what is eating your team's hours. Email is fastest, I reply within one business day.