AI Meeting Summarizer

An intelligent web application that transcribes meeting audio and generates structured, action-oriented summaries, complete with interactive AI chat capabilities.

View Live Demo »

This project was engineered to exceed the requirements of the hiring process for Unthinkable Solutions. It transforms raw meeting audio into a clear, actionable, and searchable knowledge base, demonstrating a deep understanding of full-stack AI application development.

This isn't just a summarizer; it's a complete meeting intelligence platform.

The application follows a logical, robust data processing pipeline designed for efficiency and clarity.

1. Upload: The user uploads an audio file via the Streamlit frontend.
2. Transcribe: The audio is processed by the selected Whisper model to generate an accurate text transcript.
3. Analyze: The transcript is sent to Google Gemini with a sophisticated, multi-part prompt that commands it to perform a detailed analysis.
4. Store: The resulting summary and the full transcript are saved to the SQLite database with a timestamp.
5. Display & Interact: The structured summary is presented in a clean, tabbed interface, and the user can now ask custom questions about the meeting.

Every component was chosen to meet professional standards for quality, efficiency, and scalability. This directly addresses the Technical Expectations of the project.

This project was built from the ground up to excel in the specific areas of evaluation.

• Solution: By integrating Whisper, a best-in-class ASR model, and Gemini, a top-tier LLM, the core output of the application is of the highest possible quality. The user-selectable model size for Whisper (tiny to medium) further demonstrates an understanding of the critical engineering trade-off between performance and precision.

• Solution: The project's "secret sauce" is its highly detailed, multi-part system prompt. It goes far beyond a simple request, acting as a comprehensive set of rules for the AI. This ensures consistent, high-quality, and structured output.

  Click to view the prompt's core instructions

  
  You are an expert meeting summarizer and analyst with advanced skills in understanding structured and unstructured spoken-language transcripts. Your role is to transform the following meeting transcript into a clear, insight-rich summary.
  
  ---
  ### PRIMARY ANALYTICAL INSTRUCTIONS
  Your goal is to extract maximum meaning, clarity, and actionable insight from the transcript. You will analyze the content based on the following rules before formulating your final response.
  
  **1. Understanding and Precision:**
  - Read through the entire transcript before summarizing to grasp full context.
  - Identify speaker roles and topic shifts.
  - Merge fragmented or interrupted speech into coherent ideas.
  
  **2. Summarization Style:**
  - Be concise, objective, and factual. Paraphrase in a professional tone.
  - Limit the main summary to 6–10 sentences but ensure completeness.
  
  **3. Decisions Extraction:**
  - Include only confirmed decisions, not suggestions.
  - Start each decision with a strong action verb (e.g., Approved, Finalized, Agreed).
  - If a decision is conditional, mark it as *Pending confirmation*.
  
  **4. Action Items Extraction:**
  - Each action item must include an Owner, a Task, and a Deadline.
  - If the transcript lacks an owner or deadline, infer logically from context and mark any missing info as **(TBD)**.
  - Action items must start with a verb (e.g., Prepare, Review, Submit).
  
  **5. Handling Edge Cases:**
  - Ignore conversational fillers ("uh", "you know") and off-topic digressions.
  - Merge discussions on the same topic that occur at different times into one cohesive point.
  - If critical information is missing due to poor audio, mention "Information incomplete in transcript."
  
  ---
  ### CRITICAL OUTPUT REQUIREMENT
  
  After performing the detailed analysis above, your entire output **MUST BE A SINGLE, VALID JSON OBJECT** and nothing else.
  
  - Do **NOT** use Markdown headings (like '### Summary').
  - Do **NOT** add any introductory text, explanations, or closing remarks.
  - The response must start with `{` and end with `}`.
  
  The JSON object must have the following exact structure:
  {
      "summary": "A concise but comprehensive paragraph overview of the entire meeting.",
      "key_decisions": [
          "Decision 1...",
          "Decision 2..."
      ],
      "action_items": [
          {
              "owner": "Person or Team responsible",
              "task": "The specific action to be taken.",
              "deadline": "The deadline for the task, or (TBD)."
          }
      ]
  }
  ---
  
  Now, analyze the following transcript and provide the JSON output:
  
  

• Solution: The codebase is logically partitioned into app.py for the user interface and logic.py for all backend processing (database, AI calls). This separation of concerns is a professional best practice that makes the code clean, scalable, and easy to maintain.

1. Clone the repository:

   git clone [https://github.com/sanjayjr8/ai-meeting-summarizer.git](https://github.com/sanjayjr8/ai-meeting-summarizer.git)
   cd ai-meeting-summarizer

2. Create and activate a virtual environment:

   python -m venv venv
   .\venv\Scripts\activate

3. Install dependencies:

   pip install -r requirements.txt

4. Set up your API key:

   • Create a folder: .streamlit
   • Inside it, create a file: secrets.toml
   • Add your key: GEMINI_API_KEY = "YOUR_KEY_HERE"

5. Run the app:

   streamlit run app.py