CASE STUDY — INTERNAL TOOLS + AI

Baseplate

A production operations platform built for Loadout Media: every project, asset and client approval in one place, from first brief to signed off.

CLIENT
Loadout Media
SCOPE
Product design · Build · AI integration
STACK
Angular · NestJS · Postgres · BullMQ · S3 · OpenAI · Tauri
YEAR
2026 — ongoing
Baseplate projects board: every project as a card in its lifecycle column

/01

The problem

A busy production studio ran on four spreadsheets, two cloud drives and a lot of email. Nobody could answer "what stage is this project at?" without asking three people. Footage landed in whichever drive the shooter used that day, and client feedback arrived as timestamps typed into replies.

Off-the-shelf options each solved a third of it. Asset managers ignored the project lifecycle; review tools ignored the archive; project trackers knew nothing about media. Stitching three subscriptions together still left the admin work with the producers.

6
WEEKS TO FIRST WORKING BUILD
9 HRS
PRODUCER ADMIN SAVED PER WEEK
4→1
TOOLS REPLACED BY ONE SYSTEM
2 DAYS
AVERAGE APPROVAL TURNAROUND

/02

What we built

One system organised around the project lifecycle, because that is how the studio already thinks: upcoming, in shoot, collecting assets, in edit, in review, signed off. Everything else hangs off that spine — the footage, the edit, the client's notes and the sign-off.

  • PROJECTS
  • INGEST
  • AI INDEXING
  • CLIENT REVIEW
  • SCHEDULING

FROM BRIEF TO SIGN-OFF

One spine, six stages

Every project moves along the same lifecycle, and every tool in Baseplate hangs off it. Nobody has to ask where something is up to.

  1. 01PLAN

    Brief, crew and calendar

    Projects start on the lifecycle board with a client, a type and a storage budget. Crew, studio rooms and shoot days are scheduled against real capacity.

  2. 02SHOOT

    Run sheets on the day

    Shoots carry the call time, the crew and the projects being filmed, so everyone on set knows what each card is for.

  3. 03INGEST

    Cards file themselves

    Upload in the browser, through a client upload link, at the four-lane ingest station or onto the Baseplate drive. It all lands in the right project folder.

  4. 04PROCESS

    Proxies, transcripts, tags

    The pipeline probes each file, cuts a 720p proxy, transcribes it, describes and tags it with a vision model, then indexes it for search.

  5. 05REVIEW

    Feedback on the frame

    Renders go out as deliverables (full length, shorts, trailers) behind one link per project. Clients comment on the exact frame.

  6. 06SIGN OFF

    Approvals on record

    Each deliverable is approved on its own, with a quotable record ID. When every one is signed off, the project is too.

PROJECT WORKSPACE

Every asset in its place

Each project opens on its own workspace: numbered folders for video, stills, audio, renders and documents, with proxies, durations and AI tags on every card.

  • Files route into folders by type on upload — no naming conventions to police
  • Filter by AI tag, sort by date, size or name, and flag selects for the edit
  • Camera RAW stills and pro codecs get browser previews automatically
A Baseplate project workspace with numbered folders and tagged render cards
Baseplate asset panel with an AI description, keyword tags and a speaker-labelled transcript

AI INDEXING

Search the archive by what was said

Every clip is transcribed with speaker labels, described and tagged on ingest, then embedded for semantic search. Ask for "host laughing at a guest story" and the right takes come back.

  • Speaker-labelled transcripts with timecodes
  • A written description and keyword tags for every clip
  • Director's select scores each take on focus, exposure, framing and performance
  • Semantic search across every project and shoot

INGEST

Footage files itself

Drop files anywhere and the pipeline takes over: classified on upload, routed to the chosen project or shoot, then proxied, transcribed, tagged and indexed, with every stage visible in a live queue.

  • Six-stage pipeline with live progress for every file
  • Auto-routing by type into video, stills, audio and documents
  • Public upload links for clients and freelancers, no account needed
  • Stored on our own S3 cluster, not a per-seat cloud drive
Baseplate uploads page showing the six-stage ingest pipeline and auto-routing

/03

Where AI earned its place

We only automated tasks people were already doing badly because they were dull. Three of them, each measured before it shipped.

  • Transcription and speaker labelling on ingestWHISPER
  • Clip descriptions, keyword tags and director's select scoresVISION MODEL
  • Transcript summaries and semantic search across the archiveEMBEDDINGS

Nothing auto-deletes, auto-publishes or auto-approves. Every AI output is a suggestion a human accepts — which is the reason the team trusts it.

CLIENT REVIEW

Feedback pinned to the frame

Studio and client watch the same cut. Comments land on the exact frame, drawings sit on the picture, and every version keeps its history, so nobody goes digging through an inbox for notes.

  • Frame-accurate comments, with drawing on the picture
  • Version history, with the latest cut taking the feedback
  • Live presence: see who else is watching
  • Full-quality originals when the browser can play them
Baseplate review screen with the player, comment timeline and frame-pinned comments

ALSO IN THE BOX

The rest of the kit

“Monday mornings used to be an hour of asking where things were. Now the board answers it before anyone opens Slack.”
HEAD OF PRODUCTION, LOADOUT MEDIA

/04

Where it is now

Baseplate runs every Loadout Media project and is being rolled out to two other studios. The parts that generalise — ingest routing, transcript search, frame-pinned review — are the same parts we reuse when a client needs a system of their own.

NEXT CASE STUDYDeployrINFRASTRUCTURE SOFTWARE — 2026 — ongoing

SIMILAR MESS, DIFFERENT INDUSTRY?

Let's map your process.

[email protected]