Personal projectC# / .NETApplied AI

Real-Time Speech Transcription & Knowledge Assistant

A personal Windows desktop application combining speech transcription, local document retrieval, and source-linked responses through asynchronous communication.

Request a project walkthrough

Desktop application architecture

C#
Windows Forms: .NET 4.8
Framework

Audio transcription
Local document index
Source-linked responses
Cancellation & UI updates

Implementation overview · Personal project

Source-based overview of the desktop, audio, retrieval, and response workflows.

Summary

Project context

Independently developed a Windows Forms application that connects audio processing and document search with a responsive desktop interface.

Problem / goal

Audio streams, document processing, and remote responses arrive at different speeds. The desktop application needed to coordinate them while keeping the interface responsive and supporting cancellation.

My role

Sole developer of a personal C#/.NET Framework desktop project.

My contributions

Built the Windows Forms interface and separated coordination, local retrieval, and remote client responsibilities.
Integrated NAudio/WASAPI audio capture, resampling, WebSocket transcription, and an HTTP segment fallback when connection startup fails.
Implemented PDF, DOCX, and PPTX text extraction with content hashes, incremental local indexing, and bounded retrieval of source snippets.
Connected retrieved context to streamed responses over HTTP/SSE and coordinated background tasks using cancellation, SemaphoreSlim, and UI-thread updates.

Technical approach

Keep audio, retrieval, and response clients behind separate interfaces so orchestration can manage lifecycle and cancellation.
Use source snippets and references to connect generated responses back to local documents.
Bound concurrent work and marshal UI changes onto the Windows Forms thread.

Key features

Real-time transcription with a connection fallback
Local PDF / DOCX / PPTX retrieval
Source-linked streamed responses
Cancellation and bounded background work

Impact / results

Implemented the desktop, audio, retrieval, and asynchronous communication paths in a personal application.
Source review supports the implementation described here; no measured answer-accuracy, throughput, user-count, or test-pass result is published.

What I learned

Cancellation and resource disposal matter across every stage of an asynchronous desktop workflow.
Source references make answers easier to inspect, while answer quality still requires separate evaluation.

My contributions

Built the Windows Forms interface and separated coordination, local retrieval, and remote client responsibilities.

Integrated NAudio/WASAPI audio capture, resampling, WebSocket transcription, and an HTTP segment fallback when connection startup fails.

Implemented PDF, DOCX, and PPTX text extraction with content hashes, incremental local indexing, and bounded retrieval of source snippets.

Connected retrieved context to streamed responses over HTTP/SSE and coordinated background tasks using cancellation, SemaphoreSlim, and UI-thread updates.