Internal tool · AIIn use

Intake

AI-assisted brief condenser that turns messy conversations into actionable project scopes.

Context
Internal tool & client onboarding workflow
Role
Product Architecture, AI Prompt Engineering, Frontend & API
Timeline
5 weeks · 2024
Key Stack
Next.js 14, TypeScript, Anthropic Claude API
70%
Reduction in scoping hours
2 days
Proposal turnaround cut from 7
92%
Client sign-off on first review
01 The Problem

Why this needed to be built.

Intake solves the hardest part of early client engagements: extracting clear technical requirements from open-ended conversations. It takes messy emails, voice memos, and meeting transcripts and synthesizes them into structured project briefs with explicit open questions, risks, and timeline estimates.

Prospective clients rarely come with clean software specs. They explain their pain in business terms, leaving builders to spend days going back and forth on scope definitions. Without disciplined intake, projects suffer from scope creep before the first invoice is even drafted.
02 The Solution

Approach and execution.

Built a guided pipeline powered by LLM extraction routines that parse transcripts, tag constraints, surface hidden assumptions, and draft structured proposals with milestone deliverables, timeline estimates, and acceptance criteria.

Deliverables

  • Multi-modal intake portal (audio notes, emails, rough text)
  • Structured proposal generator with scope guardrails
  • Automatic milestone and budget estimator
  • Export to Notion, PDF, and Linear
03 Highlights

Key functional details.

Voice-to-Brief Synthesis

Upload a 10-minute kickoff audio recording to produce an executive scope summary in under 45 seconds.

Assumption Spotter

Flags unstated dependencies like third-party API limits, asset readiness, or authentication requirements.

Milestone Formatter

Packages the brief into fixed-scope two-week sprint milestones with clear deliverables.

04 Technical Retrospective

What this project proved

The key to high-utility AI workflows is constrained formatting. Rather than letting the model write endless essays, forcing it into strict JSON schema output made the generated briefs immediately actionable.

Tech stack:Next.js 14TypeScriptAnthropic Claude APIWhisper AITailwind CSSSupabase