Omnilogic Labs
// CASE STUDY — FILM & TV LICENSING

A platform that reads the way they read.

A film & TV IP-licensing company acquires and licenses titles, which means reading and assessing a constant stream of scripts. Done their way, that took roughly ten hours per script — structure, character, scene efficiency, dialogue — expert judgment that doesn't scale by hiring, because the value is in their particular way of reading.

We built a script-analysis platform that runs customizable analysis modules, refines results conversationally, and learns the house method — then made that method provable, with an evaluation corpus and content-hash reproducibility underneath. A one-off engagement became a recurring product in daily use.

DOCUMENT-INTELLIGENCEMETHODOLOGY CAPTUREEVAL CORPUSREPRODUCIBLE
FIG. ONE-OFF → PRODUCT
REF: SAP-00Flywheel: a single engagement extracted into a reusable, recurring product

01 // The problem

The bottleneck was the expertise — and it lived only in people's heads.

Acquiring and licensing IP means assessing a constant inflow of scripts. Done well, a single assessment ran something like ten hours — structure, character, scene efficiency, dialogue — and it was exactly the kind of judgment that doesn't scale by hiring. The value wasn't in reading a script; it was in reading it their way.

So the work was two problems at once. It was an operational bottleneck on the business. And it was a body of proprietary method with no home outside the people who carried it. Lose the people and you lose the method. A generic model over a document would have produced analysis — just not theirs, which is the only analysis that was worth anything here.

SPEC_ID: SAP-PROBLEM
BASELINE: ~10 HRS EXPERT ANALYSIS / SCRIPT
DIMENSIONS: STRUCTURE / CHARACTER / SCENE / DIALOGUE
RISK: METHOD UNDOCUMENTED, NON-PORTABLE

02 // The approach

Don't run a model over a document. Teach the document how they read.

We didn't build a summarizer. We built a workspace over the company's own corpus, with the machinery for the company to encode its reading into the system — and the controls to keep it encoded as the system grows.

FIG. ANALYSIS PIPELINE
REF: SAP-ARCHPipeline: upload, run modules, refine conversationally, version the result
  • Customizable analysis modules — story structure, character, scene efficiency, dialogue — run independently, not a single monolithic pass.
  • Conversational refinement: results aren't a final verdict, they're a draft the reader interrogates and steers in their own language.
  • An "extract insights" capability that lets the company teach the system its own methodology, so analysis converges on how they read.
  • An evaluation corpus of reference scripts to measure analysis quality against — quality becomes measurable, not a vibe.
  • Content hashing of the analysis basis, so every analysis is versioned and reproducible — re-runnable and comparable over time.

03 // What we built

A workspace on top, a verifiable basis underneath.

REF: SAP-MOD-01
upload_file

Modular analysis

Upload a script and run the modules that matter for this title — structure, character, scene efficiency, dialogue — each producing its own assessment.

REF: SAP-MOD-02
school

Extract insights

The capability that makes it theirs: the company teaches the system its own methodology, so the analysis converges on the house way of reading — not a generic one.

REF: SAP-MOD-03
forum

Conversational refine

Results are a draft, not a verdict. The reader interrogates, corrects, and steers the analysis in their own words until it says what they mean.

REF: SAP-MOD-04
difference

Rewrite comparison

Compare drafts of the same script side by side so a rewrite can be assessed against the version that came before, not from memory.

REF: SAP-MOD-05
fact_check

Evaluation corpus

A set of reference scripts the analysis is measured against, so changes to the system can be checked for whether they help or quietly regress the house standard.

REF: SAP-MOD-06
tag

Content hashing

Every analysis is keyed to a hash of its inputs, so a result is versioned and reproducible — you can always tell what was analyzed, and re-derive it exactly.

04 // What made it hard

Capturing tacit expertise — and proving it stayed captured.

The whole point was the proprietary method, so the engineering challenge was never "run a model over a document." It was capturing tacit expertise — and then verifying it stayed encoded as the system and the corpus grew.

That is why the evaluation corpus and the analysis-basis hashing exist. They aren't plumbing; they are the proof. The corpus makes "does this still read the way we read?" a measurable question instead of an opinion. The hash makes any answer reproducible — same basis, same analysis, every time. Knowing which knob to turn meant building the instruments to tell whether turning it helped. Making the system theirs, and provably so, was the work.

SPEC_ID: SAP-MOAT
// THE INSTRUMENTS
  • Quality is measured against a corpus, not asserted.
  • Every analysis is hashed to its basis — reproducible by construction.
  • The house method is encoded in the system, not in someone's head.
// TUNE TO THE RIGHT SIGNAL — AND VERIFY IT

05 // The outcome

A one-off engagement that became a product they rely on.

REF: SAP-OUT-01

~10 HRS / SCRIPT

of manual expert analysis replaced per script by the platform.

REF: SAP-OUT-02

DAILY USE

in production and active iteration — a working tool, not a pilot left on a shelf.

REF: SAP-OUT-03

ONE-OFF → RECURRING

a single engagement grew into a recurring product the company depends on.

DOCUMENT-INTELLIGENCESCRIPT ANALYSISMETHODOLOGY CAPTUREEVALUATION CORPUSVERSIONED / REPRODUCIBLECONVERSATIONAL REFINEMENT
// NEXT

Have a method that lives only in people's heads?

If the value of your work is in a particular way of reading, judging, or deciding — and it doesn't scale by hiring — we can help you encode it into a system, and prove it stayed encoded. Names and proprietary methods are withheld by agreement; we're glad to walk through the details under NDA.

Start a conversationarrow_forwardBack to all work