AWS Builder Center
Weekend Creative Agent Challenge: IntroCraft - Autonomous Presentation Coach & Daily Speech Agent

Weekend Creative Agent Challenge: IntroCraft - Autonomous Presentation Coach & Daily Speech Agent

IntroCraft is an AI-powered presentation coach that uses AWS Bedrock to generate, score, and refine engaging 45–60 second presentation openings. Its retro Minecraft-inspired interface adds interactive teleprompter and rehearsal features to help developers, founders, researchers, and students improve delivery.

Building Tech for Impact - @TechSangi

Vision & What It Does

Public speaking and presentation delivery are two of the most critical skills for software developers, founders, researchers, and students. However, crafting a high-impact opening statement—one that captures immediate attention and establishes instant trust within the first 45 to 60 seconds—remains a major hurdle.
IntroCraft is an always-on creative agent designed to solve this problem. Operating autonomously, IntroCraft leverages 4-Part Presentation Blueprint (Hook, Credibility, Why Now, Audience Value) and AWS Bedrock AI Foundation Models to continuously generate, refine, and deliver tailored presentation openings.

Creative Output & Always-On Agent Functionality

IntroCraft functions as a persistent presentation coach:
  1. Daily Autonomous Speech Generation: On a scheduled loop, IntroCraft ingests trending developer topics, cloud architectures, and student milestones to autonomously generate fresh 4-part speech openings across four distinct rhetorical strategies (Personal Journey, Mindset Shift, Interactive Challenge, and High-Stakes AWS Reality).
  1. Quality Heuristic Scoring: Evaluates every generated speech opening across 6 core delivery metrics: Attention, Credibility, Relevance, Audience Value, Spoken Naturalness, and Memorability.
  1. Interactive Teleprompter Mode: Converts generated scripts into fullscreen rehearsal teleprompters complete with timed auto-scrolling (0.75x–1.5x speed), font scaling, and spoken delivery cues (e.g.  [Pause 2 seconds] ,  [Look up at audience] ).
  1. Retro 100% Minecraft Pixel Art UI: Designed with a nostalgic, pixelated 3D bevel interface (built using Google Fonts  VT323  and  Press Start 2P  with  @hugeicons/react ), making presentation practice engaging, memorable, and fun.

How I Built It

Developing IntroCraft required combining strict presentation coaching frameworks with robust cloud-native AI architecture.

Development Process & Key Decisions

  • Rhetorical Blueprint Structuring: Rather than generating unstructured generic paragraphs, IntroCraft decomposes every presentation introduction into four essential pillars:
  •  [HOOK] : Earns immediate audience attention in sentence one.
  •  [CREDIBILITY] : Establishes authority in a single fluid line without sounding like a resume dump.
  •  [WHY NOW] : Creates current urgency, technical shifts, or real-world stakes.
  •  [AUDIENCE VALUE] : Promises one tangible takeaway the audience will gain.
  • Iconography & Design Consistency: We opted for 100% Hugeicons integration ( @hugeicons/react  and  @hugeicons/core-free-icons ) combined with custom pixel art CSS styling to deliver a responsive Minecraft aesthetic.

Challenges Encountered & Solutions

  1. AWS Bedrock Rate Limits & Single-Region Token Quotas:
  • Challenge: Direct invocation of single-region foundation model endpoints (like  us-east-1  directly) can hit daily token bucket limits during high-frequency autonomous agent generation loops.
  • Solution: We implemented AWS Cross-Region Inference Profiles ( us.meta.llama3-3-70b-instruct-v1:0 ), allowing AWS to dynamically load-balance inference requests across multiple AWS data centers in North America. This multiplied token throughput while dramatically reducing latency. We also built an automatic multi-model fallback queue ( Llama 3.3 70B  →  Llama 3.1 8B  →  Claude 3 Haiku  →  Presentation AWS Engine ).
  1. Spoken Cadence vs Written Text:
  • Challenge: Standard LLM text outputs often sound overly formal or essay-like when read aloud.
  • Solution: We introduced spoken language constraints (contractions, active voice, 90–160 target word count) and programmed automated delivery cue tags to guide speaker pauses and eye contact.

AWS Services Used / Architecture Overview

IntroCraft is deployed using AWS Free Tier services and AWS Bedrock foundation models:

AWS Services Summary
  • Amazon Bedrock: Serves as the core AI generation engine, utilizing cross-region inference profiles ( us.meta.llama3-3-70b-instruct-v1:0  and  us.meta.llama3-1-8b-instruct-v1:0 ) to generate creative spoken openings.
  • Amazon S3 (Simple Storage Service): Hosts the production static web app ( cloudblueprint-frontend-013131247228 ) configured for static website hosting in  us-east-1 .
  • AWS CLI (Profile:  cloudblueprint ): Used for automated deployment pipelines and AWS Bedrock model credential verification.

What I Learned

Participating in the AWS Builder Center Weekend Challenge provided several valuable technical insights:
  • Power of Cross-Region Inference: Transitioning from single-region Bedrock model IDs to Cross-Region Inference Profiles ( us.meta... ) solved rate-limiting bottlenecks and provided high availability for autonomous agent workloads.
  • Prompt Engineering for Spoken Communication: Engineering prompts for spoken delivery requires distinct constraints compared to written copy—focusing on rhythm, sentence brevity, and explicit pause markers.
  • Merging Retro Gamification with Developer Tools: Combining a 100% Minecraft pixel GUI with professional speech coaching created an approachable, highly engaging user experience.

Link to App & GitHub Repository

Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article