Work
Selected work
A sample of recent engagements and solo-built products spanning sovereign AI, applied ML, real-estate technology, and enterprise data platforms.
Sovereign AI & document intelligence
EU enterprise · 2025–2026 · production RAG build + research paper
CARAG — Compliance-Aware RAG over a 1.2M-document enterprise corpus
An enterprise client needed retrieval-augmented generation over a 1.2M-document internal corpus where retrieval-time eligibility — who is allowed to see what, for which purpose — matters as much as relevance. Off-the-shelf RAG breaks here: the most relevant passage may be the most legally inadmissible.
European long/short hedge fund · 2025 · regulated-document RAG
Financial RAG for a European hedge fund
A multi-strategy European hedge fund needed retrieval-augmented research over regulated financial filings, broker reports, internal models, and news. PMs were spending hours per name on document-trawling that should be a 30-second query, while compliance demanded a defensible audit trail for every output.
B2B mid-market client · 2025 · internal RAG copilot
Internal SOP & HR copilot
A mid-market client had hundreds of pages of standard operating procedures, HR policies, and internal tooling docs scattered across Confluence, Notion, Google Docs, and Slack pinned messages. Onboarding took weeks. Routine policy questions burnt hours of manager time. Existing intranet search returned matches but not answers.
AI applications
Entertainment-industry client · 2024–2025 · multimodal media analytics + production rollout
Media analytics & creative intelligence for an entertainment client
An entertainment-industry client produced large volumes of creative assets and promotional media every quarter and needed a way to predict which ones would resonate with their audience before committing distribution and paid-media spend. Existing analytics surfaced post-launch performance but offered no signal during the pre-launch window where decisions actually get made.
Entertainment-industry client · 2025 · email classification + routing
LLM email triage for entertainment industry
An entertainment-industry company was drowning in inbound email — agent submissions, press requests, fan/PR escalations, legal notices, internal team threads — all flowing into shared inboxes with no time to triage cleanly. Important threads slipped through; routine ones consumed disproportionate attention.
Luxury fashion / accessories brand · 2025 · multimodal recommendation engine
AI-based recommendation system for a luxury brand
A luxury brand needed a recommendation system, but standard collaborative filtering breaks for luxury — purchase frequency is low, statistical co-occurrence between customers is sparse, and brand aesthetic matters far more than price clustering. A customer who buys one piece does not follow the same pattern as the next.
International NGO · 2025 · LLM-driven email triage + response drafting
NGO — AI email management for the communications team
An NGO communications team was processing tens of thousands of inbound emails — donor enquiries, partner requests, press, volunteer applications, advocacy correspondence — all flowing into shared inboxes. The team had a small full-time staff who were spending most of their day routing and drafting initial responses to repetitive enquiries.
Real estate & PropTech
US-based PropTech marketplace · 2024 · ML pipeline build
Foreclosure sale-price prediction for a distressed-asset marketplace
A distressed-property marketplace needed to estimate sale prices for foreclosure listings before they hit auction. Existing AVM tools handle conventional residential well but break down on distressed assets, where condition, occupancy status, and local market thinness matter more than comparables.
EU-based real-estate development SaaS founder · 2025 · 6-week MVP build
Real-estate development SaaS — Monte Carlo simulation engine MVP
A real-estate development SaaS founder needed an MVP to help developers and investors stress-test their project economics. Spreadsheet-based deal models break under uncertainty — a single IRR number does not tell you what happens when interest rates move 200 bps, refinance windows slip, or hard costs come in 15% over plan.
EU-based construction-tech client · 2024 · vision pipeline + causal modelling
Construction site reconciliation — computer vision + causal inference
A construction-tech company needed to reconcile 2D site sketches with actual 3D project state, then identify which of ~7,000 reported features actually drove schedule variance. A reporting tool was producing dashboards nobody could act on — "the schedule is slipping" without isolating the why.
Applied ML & data science
Climate-finance research client · 2025 · econometric impact analysis
Climate finance — policy impact analysis
A climate-finance research organisation needed to quantify the impact of specific policy interventions on capital flows and emissions — separating the policy signal from confounding macro factors, sector trends, and pre-existing trajectories.
Financial-services client · 2023 · real-time inference + monitoring
Real-time fraud detection pipeline
A financial-services client needed real-time fraud detection on transaction streams — flag suspicious activity within milliseconds of a transaction, with low false positives that would otherwise burn customer-experience budget on legitimate users.
Enterprise client (Upwork Enterprise) · 2021 · ETL + ML model deployment
AWS Enterprise ETL + ML on SageMaker
A US-based enterprise client needed a cloud-native ETL and ML platform on AWS — ingest raw data from multiple operational sources, transform it into model-ready features, train and deploy production models on SageMaker, and serve predictions back to downstream consumers with proper observability.
Agri-tech / SMB lending client · 2025 · invoice-data ML pipeline
Agri-credit scoring from invoice data
An agri-tech lender needed to extend credit to SMB farmers and food producers without traditional credit-bureau coverage. The available signal was raw invoice data — purchase orders, payment histories, supplier relationships, seasonality patterns — sitting in unstructured PDFs and accounting exports across the borrower's existing book.
Mid-market e-commerce client · 2024 · churn modelling + retention pipeline
E-commerce churn analysis & predictive retention
A mid-market e-commerce client had rising customer-acquisition costs but no reliable signal on which customers were about to lapse. Standard "haven't bought in 90 days" rules flagged everyone too late — by then the customer had already moved to a competitor.
US local political campaign · 2024 · precinct-level data analysis
County-level voter analysis for a US local political campaign
A US local political campaign at the county level needed precinct-level targeting — where to door-knock, where to spend on local ads, which voter segments to focus on. The campaign had voter-file data, demographic snapshots, and turnout history but no one to turn it into operating decisions.
International NGO · 2025 · NLP analysis on free-text survey responses
NGO — survey response analysis with NLP
An NGO ran a large-scale survey across multiple regions and received tens of thousands of free-text responses to open-ended questions. The communications and policy teams needed to surface themes, sentiment, and regional variation — but had been doing it by spreadsheet for previous waves and missing patterns in the long tail.
Solo-built products
Solo-built product · 2024–2026 · live in production
AskVault — RAG-first knowledge retrieval (solo-built SaaS)
Existing knowledge-management tools either index documents without understanding them or generate confidently wrong answers. Teams building real workflows on internal knowledge need neither — they need calibrated retrieval that says "I do not know" when the corpus does not actually contain the answer.
Solo-built product · live at goprivox.com
Privox — privacy-focused micro-SaaS
Most consumer-facing SaaS treats privacy as a checkbox. The micro-SaaS opportunity was to build a tool where privacy is the actual product mechanic, not a marketing claim — minimal data collection, no third-party tracking, clean opt-in for everything.
Want a similar engagement on your stack?
Most engagements above started with a 60-minute scoping call.