Meeting Intelligence Platform

At a Glance

• Client: Management consulting firm
• Challenge: Consultants spending evenings and weekends catching up on meeting notes instead of client work
• Timeline: Production-ready platform delivered in 4 weeks
• Solution: AI-powered transcription, summarization, and follow-up drafting — integrated into their existing intranet
• Outcome: Platform delivered and approved, ready for organization-wide rollout

A management consulting firm where consultants spend their days in client meetings — discovery sessions, strategy workshops, progress reviews, stakeholder alignments. Valuable conversations that drive their business.

The problem wasn't the meetings. It was everything that came after.

After every client meeting, consultants faced the same routine:

• Transcribe the recording (or rely on memory)
• Summarize 60-90 minutes into digestible key points
• Extract action items and assign owners
• Draft a follow-up email to send to the client

This documentation work was eating into evenings and weekends. Follow-up emails went out days late. Summary quality varied wildly depending on who wrote them. And valuable insights from meetings were getting lost in unstructured notes that nobody could search.

The consultants were frustrated. The partners knew they were losing billable hours to admin work. Something had to change.

The solution needed to:

• Integrate with their existing CMS-based intranet — no new systems for IT to manage, no new interfaces for consultants to learn
• Handle long recordings — some strategy sessions run 2+ hours
• Show progress in real-time — consultants wanted to see processing status, not wonder if it was working
• Maintain confidentiality — client discussions are sensitive; security was non-negotiable

The timeline was aggressive — 4 weeks to deliver a working platform. The firm had commitments to meet and couldn't wait months for a solution.

We built a platform that lets consultants either upload a recording or record meetings directly in the browser. Either way, they get back:

• Full transcript with speaker identification ("who said what")
• Executive summary highlighting key points and decisions
• Action items with owners extracted automatically
• Draft follow-up email ready to review and send

The live recording feature means consultants don't need separate recording software — they can start a meeting, hit record in their intranet, and when the meeting ends, processing begins automatically.

All of this accessible from their existing intranet — no new systems to learn.

How It Works

Consultant uploads recording
          │
          ▼
┌───────────────────────────────────────────────────────┐
│  CMS Intranet → API → Background Workers    │
│       │                     │                │
│       │              ┌──────┴──────┐           │
│       │              ▼              ▼           │
│       │         AssemblyAI    OpenAI GPT-4   │
│       │        (transcribe)  (summarize)     │
│       │              │              │           │
│       ◀──────────────┴──────────────┘           │
│       │                                       │
│  Real-time progress updates via SSE          │
└───────────────────────────────────────────────────────┘
          │
          ▼
Transcript + Summary + Actions + Draft Email
        (typically 5-8 minutes)

Consultants see progress in real-time as the meeting is processed. When it's done, they review the outputs, make any edits, and send the follow-up — all within the same interface they already use daily.

The Technical Approach

We designed a microservices architecture that separates concerns:

• CMS Module (PHP) — Upload interface, meeting history, user management. Lives inside their existing intranet.
• FastAPI Service (Python) — Handles API requests, authentication, job orchestration.
• Celery Workers (Python) — Background processing pipeline for transcription and summarization.
• SSE Service (Go) — Real-time progress updates pushed to the browser.
• AI Services — AssemblyAI for transcription (with speaker diarization), OpenAI GPT-4 for summarization.

The split architecture meant we could optimize each piece independently and scale processing capacity without affecting the user-facing interface.

The 4-week timeline required focused execution:

Week 1: Foundation

• FastAPI service scaffolding
• Database schema design
• Authentication system (JWT + RSA for CMS trust)
• Docker development environment
• API contract agreed with client IT

Week 2: AI Integration

• AssemblyAI transcription pipeline
• OpenAI summarization prompts (tuned for consulting context)
• Celery job orchestration
• Error handling and retry logic

Week 3: Real-time & CMS Integration

• Go SSE service for progress updates
• CMS module development (PHP)
• Progress tracking UI
• PDF report generation

Week 4: Integration & Handoff

• End-to-end testing
• CMS integration testing with client IT
• Documentation and deployment guides
• Client sign-off

Security

Given the sensitivity of client meeting content, security was built in from the start:

• RSA-signed requests between CMS and API — the CMS holds a private key, the API verifies with the public key. No shared secrets.
• TLS everywhere — all traffic encrypted in transit.
• AES-256 encryption at rest — meeting recordings and transcripts encrypted in storage.
• Role-based access — consultants only see their own meetings.
• Audit logging — every action tracked for compliance.

We delivered a production-ready platform within the 4-week timeline. The client approved the system and it's now ready for organization-wide deployment.

For the client:

• Time recovery — Consultants no longer spend hours on manual transcription, summarization, and follow-up drafting. The platform handles it automatically.
• Consistency — Every meeting summary follows the same professional format, regardless of which consultant ran the meeting.
• Faster follow-up — Follow-up emails can be generated and sent within minutes of a meeting ending, not days later.
• Searchable archive — All meeting transcripts and summaries are stored and searchable, building institutional knowledge over time.

For their clients:

• Professional communication — Consistent, well-structured meeting recaps.
• Clear accountability — Action items with owners explicitly captured and communicated.
• Faster turnaround — Documentation delivered while the meeting is still fresh.

The 4-week timeline required smart architectural decisions:

• Microservices from day one — The Go SSE service could be optimized independently without affecting the Python API or Celery workers.
• RSA-based trust — Server-to-server authentication between CMS and API avoided the complexity of managing user sessions across systems.
• Celery for async processing — Meeting processing is inherently async (minutes, not milliseconds). Celery's retry logic and dead letter queues handle failures gracefully.
• SSE over WebSockets — For one-way progress updates, Server-Sent Events are simpler and have better browser support than WebSockets.
• OpenAPI contract first — Defining the API spec upfront allowed CMS integration work to proceed in parallel with backend development.

The platform is now in the client's hands, ready for rollout to their consulting team.

We specialize in building AI-integrated platforms that connect modern capabilities with existing enterprise systems. Whether you're dealing with document processing, meeting automation, or custom AI workflows — if you have a manual process that could benefit from intelligent automation, we can help.

Our strengths:

• AI/ML Integration — Transcription, summarization, NLP, document processing
• Enterprise Integration — CMS platforms, existing infrastructure, legacy systems
• API Design & Development — RESTful APIs, real-time streaming, async processing
• Full-stack Development — Python, Go, PHP, JavaScript

Let's Talk

If your team is spending hours on manual documentation, or you're looking to integrate AI capabilities into your existing systems, we'd love to hear from you.

[Hire Me on Upwork →] | [Get in Touch Directly →]

The work described here formed part of a larger project delivered for a client. Details have been anonymised to protect confidentiality, but the technical implementation, results, and process reflect real work done.