104
VISION AND AUDIO INTEGRATION
Vision
Law firms accumulate vast volumes of information
in visual formats: scanned documents, handwritten
notes, court filings received as images, and legacy
physical records. PC Chat's vision integration pro-
cesses these files through an OCR pipeline, including
sophisticated handwriting recognition to extract and
index their content. The extracted text then flows
through the standard Kernel Memory RAG pipeline for
embedding and storage, making previously inacces-
sible visual content fully queryable through natural
language.
Audio
For audio files, PC Chat uses GPT-Transcribe (an OpenAI
speech-to-text model) in place of the OCR handler. Tran-
scribed content is then processed through the same pipe-
line as other document types. This capability transforms
PC Chat into a fully conversational assistant, supporting
hands-free interaction for attorneys reviewing documents,
in meetings, or otherwise occupied.
INTEGRATION WITH
OTHER DATA SOURCES
VIA DENODO
The firm uses Denodo as its data
virtualization platform, enabling
seamless access to multiple het-
erogeneous data sources through
a single logical layer. To connect
these data assets to PC Chat, we
leverage the Denodo AI SDK to
develop APIs, which are then con-
sumed through Semantic Kernel
plugins, the same pattern used
for all other integrations. A key
implementation detail was con-
figuring AI-specific instructions in
both the PC Chat (OpenAI model)
configuration and on the Denodo
side to ensure correct interaction
with exposed data and APIs. As
Denodo continues to evolve, these
configuration steps are expected
to become more streamlined.
