Production legal AI

Legal AI Intake

Bring first-contact information into one record the intake team can review.

An inquiry rarely arrives as a complete file. A caller explains the situation, an email adds details and attachments provide another part of the picture. I built the intake experience and supporting services that bring those pieces together, organize the information and prepare the next step for the responsible reviewer.

Book a call Follow one inquiry

Part of my legal software delivery for different law firms over the last three years. Client identities are withheld under NDA; reconstructed screens and examples use fictional data.

Last updated November 5, 2026

Intake review · Demo workspaceIllustrative

New inquiry · Demo Client A

Call transcriptEmail2 attachments

Needs review

Organized intake details

Caller asks about a new matter and emails two documents. A key date was extracted from them.

Checks

Conflict review waiting for the responsible attorney
2 documents classified: incident report, invoice
1 requested record not yet received

Suggested next step

AI suggestion

Request the missing record from the client.

ApproveEditReturn

Illustrative review screen · fictional data

From first contact to review

What the intake team receives

Before this workflow, intake context arrived across conversations, email and documents. The delivered workflow connects it to a reviewable record: the details captured, documents received, information still needed and a proposed follow-up.

The reviewer can approve the next step, edit it or return it for more work. Conflict review remains an attorney decision. The software supports preparation and routing; it does not decide whether the firm should take on a client.

Capture

Bring calls, emails and attachments into one intake record.

Prepare

Extract key details, classify documents and flag missing information.

Review

Give the responsible person the context and proposed next action.

See the handoff

Follow one inquiry

Incoming information becomes an organized record, missing material becomes visible and a proposed request goes to a person for review. This reconstruction explains the delivered workflow without exposing a client workspace.

Select a step to explore. Explore at your own pace.

Step 5 of 5

Execute and record the approved action

The approved follow-up runs through a controlled tool. The log records the approval, action and resulting workflow state.

Technical detail

Human approval precedes consequential execution. Controlled tools and action-level audit trails make the approved action and its execution reviewable, alongside workflow state.

Controlled tools
Audit trail
Workflow state

Audit trail · Demo Client AIllustrative

09:12 · Inquiry captured
Intake workflow
09:14 · Intake details and documents organized
AI assistance
09:15 · Next step suggested
AI assistance
09:20 · Conflict review completed
Responsible attorney (demo)
09:31 · Follow-up approved
Case manager (demo)
09:31 · Approved follow-up carried out
Controlled tool

Workflow state: Awaiting client reply

The inquiry, documents, names and times shown here are fictional.

Delivered capabilities

What the team can do, and what supports it

The operational task and its engineering responsibility belong together. Each capability supports the handoff from first contact to a reviewable inquiry.

Assemble the inquiry

For the intake team: Bring the call, email and attachments together so the intake team can inspect the full context.
Engineering behind it: Voice and email capture, Microsoft Graph integration and a structured intake record.

Understand the documents

For the intake team: See which documents arrived and which information still needs to be requested.
Engineering behind it: Document extraction, classification and background processing supported by Node.js and Python/FastAPI services.

Route the decision

For the intake team: Give the responsible attorney or case manager the context needed to review a proposed next step.
Engineering behind it: Review queues, permission-aware access and a boundary between recommendations and consequential actions.

Keep the action record

For the intake team: Inspect what was reviewed, who made the decision and what action followed.
Engineering behind it: Controlled execution and action-level auditability across the integration handoff.

My contribution

The application behind the intake record

My contribution covered the full-stack intake interface, workflow logic, processing services, integrations and approval controls. Document understanding, extraction and classification turned incoming files into usable context.

Intake experience
The full-stack interface and services supporting the workflow.
AI orchestration
Preparation and suggested next steps with explicit human review boundaries.
Processing services
Node.js and Python/FastAPI services, including background document processing.
Integrations
Microsoft Graph, SharePoint and supporting APIs for permitted email and document context.
Controls
Permissions, approval boundaries, controlled actions and action-level auditability.

Node.js
Python
FastAPI
Microsoft Graph
SharePoint
Background processing

Architecture

How the system connects

Intake and document services prepare the record. Review controls govern the next action, and controlled tools execute within the application's permissions.

Microsoft Graph and SharePoint connect permitted email and document context. Permissions determine which sources and actions are available to the workflow.

Explore processing and integrations

Stage 1: Input channels
Where inquiries arrive
- Voice
- Email
- Documents
- Real-time transcription
- Microsoft Graph
Stage 2: Workflow / orchestration
Organizes the work
- Intake
- AI assistance
- Background processing
- Node.js
- Python / FastAPI
Stage 3: Context
What the AI and reviewers draw on
- Case data
- Documents
- Knowledge / retrieval
- SharePoint
- Permission-aware access
Stage 4: Human governance
People decide
- Permissions
- Review
- Approval
Human decision point
- Attorney review
- Case-manager review
Stage 5: Controlled execution
Only approved actions run
- Approved actions
- Notifications / updates
- Controlled tools
Stage 6: Auditability
Actions and reviews stay on the record
- Action logs
- Workflow state
- Review trail
- Action-level audit trail

System architecture

Information flow

Where people decide

Processing and connections

Node.js and Python/FastAPI support the application and document processing.
Large files run in background jobs; scheduled and priority workflows coordinate work.
Microsoft Graph and SharePoint connect permitted email and document context.

Approval controls

Preparation does not authorize an action

A suggested next step waits for the required review and permissions before it can have an effect.

AI prepares

Transcribes calls, organizes documents and suggests the next action.

People decide

Attorneys handle conflict decisions. The responsible attorney or case manager reviews the proposed next action.

System records

Controlled tools execute within permissions; the audit trail records the review and action.

Fictional approval-record specimen

What one approved action can record

A reading example of the review boundary, not a production log or an exact product schema. All details below are fictional.

Source context: Demo intake record, call transcript and two received attachments.
Proposed action: Send a follow-up requesting the missing record.
Prepared draft: Example follow-up, version 1, linked to the intake record.
Reviewer: Case manager (fictional reviewer). Attorney handles conflict review separately.
Decision: Approve this follow-up only. This does not accept the case.
Permission check: The messaging tool is permitted to contact this inquiry's recorded recipient.
Execution record: Approved follow-up sent by the controlled tool; reviewer and action recorded.
Resulting state: Awaiting the client's reply. Case acceptance remains an attorney decision.

Delivery and continued improvement

The work continues beyond the first release

This is production workflow engineering: the application people use, the services that process information and the integrations that connect the work. Continued improvement is part of my delivery history as client requirements and operational needs develop.

Clear review responsibilities

The workflow distinguishes preparation, a person's decision and the action that follows. Permission and approval boundaries are part of the application behavior.

Data governance in the workflow

Permitted email and document context supports the inquiry. Access scope and the action record connect the information used to the decision being reviewed.

Scoped secure delivery

Permission-aware access, review boundaries and action records make the controls in this workflow inspectable. This case describes that implemented scope without implying a certification or an independent security assessment.

Evidence and limits

What this case demonstrates

Implemented: A production intake workflow connecting information capture, AI-assisted preparation, human review and controlled execution.
My contribution: The intake experience, workflow logic, processing services, integrations and controls described above.
Illustrated: A reconstruction of the workflow using fictional data so client-sensitive details remain abstracted.

Limits. This case does not establish quantified time savings or an independent security assessment. Other capabilities reflect separate client work, mostly under NDA.

Explain the technical terms

Intake: The first steps after a potential client gets in touch: capturing their details, understanding the request and deciding what happens next.
Conflict review: Checking that taking on a new client will not clash with the firm’s duties to existing or former clients. A person makes the decision.
Document intelligence: Software that reads documents to understand them, pick out key details and sort them by type.
Workflow orchestration: Coordinating the steps of a process: what runs next, who reviews it and what happens after approval.
Permission-aware access: The AI system can only see the information that the workflow’s permissions allow.
Human-in-the-loop: A person must review and approve before the system takes a consequential action.
Audit trail: A record of each approved action, who reviewed it and how it was carried out.
Consequential action: A step with real effect, such as a notification or an update, that needs a person’s approval first.
Controlled tool: A defined action the system carries out only after approval and within its permissions.
Knowledge / retrieval: Finding relevant stored information to give the AI system and reviewers context.
Background processing: Heavy work, such as processing large files, runs separately so the application stays responsive.
Microsoft Graph and SharePoint: Microsoft Graph is Microsoft 365’s interface to email and files; SharePoint stores documents. The workflow connects to both.

Your intake process

Bring one inquiry from first contact to review

Where does the information arrive, what is missing and who decides the next step? Bring those questions and the systems involved to an intake workflow discussion.

Book a call Email Hamza