A few projects we can talk about.

Most of our work is under NDA. Except where a partner has agreed to be named, the summaries below are anonymized at the customer's request — domain, problem, approach, and outcome only.

  1. Named client

    JUO Company · Pet platform

    One identity across commerce, ERP, and AI travel

    As the business grew into a pet-adoption franchise, pet-care commerce, and pet travel, we pulled every service's separate member table into one central identity system.

    on one account
    3 products
    auth run in-house
    No external IdP
    a repo per service
    Independent deploys
    Read the case

  2. Anonymized · Pharma R&D

    Regulatory document AI for a pharma R&D unit

    Problem

    Regulatory dossiers were assembled manually from hundreds of source documents — PDFs of varying vintages, spreadsheets, scanned forms — with the same data re-entered three or four times before submission. Reviewers spent more time chasing inconsistencies than evaluating the science.

    Approach

    We built a multimodal document pipeline that classified and parsed source artifacts, then surfaced extracted fields in the customer's existing review tool. Each extraction carried a confidence score, a link back to the source pixel region, and a human-in-the-loop checkpoint. The eval harness ran on every model update before promotion.

    Outcome

    Submission preparation time reduced materially; reviewer time shifted from data entry to scientific judgment. The pipeline runs on the customer's Korean cloud infrastructure with full lineage from source PDF to submitted artifact, satisfying the FDA-EMA guidance principles for AI in regulatory decision-making.

  3. Anonymized · Consumer platform

    Multimodal product & condition classifier for a consumer platform

    Problem

    User-submitted images and free-text descriptions arrived in a torrent. Manual triage to product categories and condition flags was the bottleneck blocking growth and a degraded user experience.

    Approach

    A lightweight multimodal model fused image features with text descriptors, returning a category, a condition tag, and a routing decision in under a hundred milliseconds. The model lived behind an internal API; we shipped the eval harness so the customer's team could retrain on new data without us.

    Outcome

    Triage latency collapsed from hours to seconds; the operations team was redeployed from labeling to exception handling. The customer now owns and runs the model.

  4. Anonymized · Travel operations

    Operations automation for a premium tour operator

    Problem

    Itineraries, vendor confirmations, and guest preferences were spread across email threads, spreadsheets, and a legacy reservation system. The team was excellent; the tooling was not.

    Approach

    We replaced the brittle middle layer with a typed pipeline: emails parsed into structured events, vendor side-channels normalized, and a single source of truth synchronized with the legacy system. Where an LLM made the call, we logged the input, the output, and the override path for the human operator.

    Outcome

    Per-trip ops time dropped sharply; the same team now handles a much larger book without adding headcount. Audit trails per booking are exportable on demand.

Start a conversation.

Most of our engagements begin with a 30-minute call. We'll tell you honestly whether we're the right fit.

Talk to us hello@tenetcode.com