Measurable outcomes from live deployments.

The Challenge

A mid-sized Hong Kong lender was processing 80–120 loan application files per day. Each file required a staff member to open, read, classify, extract key fields, and route to underwriting. The process took 3–4 minutes per page across an average 6-page file — consuming nearly 4 full-time hours of capacity daily.

What We Built

DeView built a document automation pipeline: inbound PDFs and scanned forms are processed through an extraction layer that classifies document type, pulls structured fields (applicant name, income figures, LTV ratios, employer details), and routes to the correct underwriting queue. Low-confidence outputs are flagged for human review; everything else moves automatically.

The Challenge

A regional insurance company's support team was handling 600–800 enquiries per day across email and web chat — primarily policy questions, claim status checks, and renewal reminders. Agents spent 8–12 minutes per interaction manually searching policy documents, looking up account records, and writing responses from scratch.

What We Built

DeView built a support AI connected to the company's policy knowledge base and CRM. For each incoming enquiry, the system classifies intent, retrieves the relevant policy section and account context, and drafts a response ready for agent review. Escalations are automatically routed with full context pre-loaded — the agent sees the customer history, the issue category, and the AI's draft before picking up the conversation.

The Challenge

A professional services firm with 140 employees had 8 years of internal documentation spread across Confluence, SharePoint, and shared drives — procedures, client methodologies, compliance frameworks, and project templates. New hires took 2–3 weeks to become productive. Senior staff were interrupted 12–18 times per day answering internal questions that existing documentation already covered.

What We Built

DeView built a permission-controlled internal knowledge assistant that ingests and indexes the firm's entire document library. Staff ask questions in natural language; the assistant searches across all connected sources and returns cited answers pointing to the specific document section. Access is scoped by role — junior staff cannot retrieve materials outside their clearance level.

The Challenge

A Hong Kong financial services company produced a weekly management pack covering pipeline performance, risk flags, and market commentary. Four team members spent a combined 9 hours each week pulling data from five different systems, formatting it into a standard template, and writing narrative summaries. The report was frequently late and occasionally contained reconciliation errors.

What We Built

DeView connected the AI to the company's SQL databases, BI tools, and internal data feeds via read-only service accounts. A scheduled job runs each Friday morning: the system pulls the prior week's data, runs the standard calculations, generates the narrative summaries using the company's approved commentary format, and delivers a complete first-draft report. The team reviews, adjusts commentary as needed, and signs off.

The Challenge

IRVA coordinates post-disaster building reconstruction across developing regions, managing 15–20 active sites with 400+ registered volunteer engineers and architects. Field teams submitted damage assessments, progress photos, and material requests via WhatsApp and email — nothing structured, nothing searchable. A 3-person HQ team spent 4+ hours daily copy-pasting updates into spreadsheets. Donor reports for 12 active grants — each with different KPI formats — took one coordinator 10–12 days per reporting cycle. Volunteer-to-project matching relied on memory and a shared Google Sheet.

What We Built

DeView ran a workflow audit to map IRVA's coordination pipeline end-to-end, then built three integrated systems. First, a document automation layer that ingests WhatsApp messages and email attachments, classifies them by site and report type (damage assessment, progress update, material request, safety incident), extracts structured data, and writes it into a central project tracker. Second, a reporting copilot connected to the project database and each donor's reporting template — generating grant reports in the required format with one click. Third, a knowledge assistant indexing building codes, safety standards, and IRVA's construction playbooks across 30+ countries — so volunteers can ask 'What is the seismic foundation requirement for a two-storey school in Nepal?' and get a cited, current answer.

The Challenge

PacificPort Logistics handles 200+ ocean and air freight shipments daily through Hong Kong, connecting manufacturers across southern China with buyers in 40+ destination markets. Each shipment generates 6–12 documents — bills of lading, commercial invoices, packing lists, certificates of origin, customs declarations, and carrier manifests. Operations staff manually keyed data from each document into the TMS, cross-referencing tariff codes and compliance requirements. The process consumed 45 minutes per shipment and was the single largest bottleneck in the operation. Clients called or emailed 80–120 times per day asking for shipment status, ETAs, and document copies.

What We Built

DeView deployed four connected systems. A document automation pipeline ingests all inbound shipping documents — PDF, scanned paper, and EDI — classifies each by document type, extracts structured fields (consignee, HS codes, weights, values, port pairs), validates against customs requirements for the destination market, and writes clean data directly into PacificPort's TMS. A customer support assistant handles inbound client enquiries — connected to live tracking data, it answers shipment status questions with real-time ETAs and links to relevant documents. A knowledge assistant indexes customs regulations, tariff schedules, and restricted-goods lists across 40+ trade lanes so operations staff get instant cited answers instead of searching government portals. A daily reporting copilot auto-generates the operations dashboard, exception alerts, and weekly carrier performance reports.

The Challenge

Ashford & Lam is a 45-lawyer commercial law firm in Hong Kong with practices in M&A, banking & finance, and corporate advisory. During due diligence, junior associates spent 60% of their billable time reading contracts — manually extracting key clauses, flagging non-standard terms, and building comparison matrices across stacks of 50–200 documents per deal. A single mid-market transaction required 2–3 full days of document review before a partner could form a preliminary risk assessment. The firm's 8 years of precedent opinions, memos, and template clauses lived across a shared drive with no search beyond filename — partners relied on memory to recall relevant past work. Client billing narratives and weekly matter updates consumed another 3–4 hours per partner per week.

What We Built

DeView built four integrated tools for the firm. A document automation pipeline that ingests deal documents (share purchase agreements, shareholders' agreements, loan facilities, leases), extracts key commercial terms and obligations, flags non-standard or missing clauses against the firm's own playbook, and generates a structured comparison matrix that a senior associate can review in hours instead of days. An internal knowledge assistant indexes all precedent work — past opinions, negotiation memos, template clauses, and regulatory guidance — so any lawyer can search for relevant past work and get cited results with document links. A reporting copilot that auto-generates weekly client updates and billing narratives by pulling from time entries, milestone data, and matter status notes. A client intake system that classifies new enquiries by practice area, runs preliminary conflict checks, and routes to the appropriate partner with a structured brief.