Solutions Architect @ Amazon Web Services

Cameron Keene

I help builders architect and deploy AI-powered applications on AWS. Focused on generative AI, Amazon Bedrock, and voice AI — from re:Invent stages to production.

Career

Experience

University of FloridaB.S. Computer Science, Dec 2022
  1. Amazon Web Services

    Solutions Architect

    Jan 2025 – Present

    Dallas, TX

    • Poor customer experience costs businesses an estimated $3.7T annually — co-developed and co-presented a re:Invent 2025 breakout session on intelligent contact center solutions to address this; architected a multi-channel agentic platform using Amazon Bedrock with a partner TTS/STT provider, LangChain SDK, and TypeScript, delivering reference architectures that normalize SMS, email, text, and voice into a unified message stream with custom session state management, incorporating real-time AI flows, autonomous agents, and human augmentation; session drew 71 in-person attendees and produced a published open-source reference implementation on aws-samples.
    • Fortune 500 sales teams lacked automated AI-driven lead discovery and qualification at scale — founded and own SONAR, an internal agentic AI sales platform, engineering a multi-agent linear system with four specialized agents: organizing net-new data, generating leads, formulating and persisting outputs, and comparing results against Salesforce to detect overlap and sync qualified leads; deployed to 7 accounts with planned rollout to 24, generating $1M ARR in new pipeline during an 8-week pilot and scaling to ~300 additional qualified leads over 7+ months; presented to L8/L10 leadership, resulting in adoption by demand gen reps and ongoing reviews with AEs/SAs driving customer adoption and strategic expansion.
    • A major enterprise customer with 120,000 monthly active users needed a fast, accurate, and scalable ban appeal processing system — designed, developed, and launched a Ban Appeal agent using AWS AgentCore primitives; implemented an orchestrator agent that defers to a specialized appeal agent or KB lookup via RAG, with tool calling through AgentCore gateway for DB lookups, Bedrock guardrails for safety, OTEL traces in Strands SDK via AgentCore observability, AgentCore memory for session management, and AWS Nova Sonic S2S model for sub-second latency on AgentCore runtime; successfully piloted to 12,000 MAUs (10% of the user base), delivering fast, accurate, and scalable ban appeal processing in production.
    • Telco customers lacked hands-on exposure to AWS AgentCore's agentic primitives, slowing adoption — developed and published a comprehensive workshop covering all AgentCore primitives (runtime, gateway, identity, memory, etc.) and delivered it twice to strong participation; accelerated customer adoption of agentic patterns and directly led into customer-run telco hackathons.
    • Major enterprise accounts required deep technical enablement and ongoing architectural guidance to successfully adopt AI workloads — led technical deep dives, workshops, and bi-weekly office hours sessions; prepared custom materials, ran hands-on sessions for senior leaders, drove customer-initiated POCs, and strengthened partner ecosystems through solution alignment and technical evangelism.
    Amazon BedrockAgentCoreLangChainStrands SDKTypeScriptNova Sonic
  2. Texas Instruments

    Software Developer

    Jan 2023 – Jan 2025

    Dallas, TX

    • TI relied on a third-party credit lending provider, creating cost and customization constraints — designed and developed an internal replacement platform, architecting the full relational schema (ER diagrams, tables, relationships, constraints), JSON schemas, API integrations, and auto-approval workflows end-to-end; resulted in approximately $250M in customer credit extended through the new platform.
    • TI.COM suffered 10–15 second page load times due to real-time DB queries serving batch-style data, degrading customer experience and limiting update frequency — architected a new batch processing layer that removed legacy database integrations, converted the frontend service to fetch data via SODA instead of real-time REST calls, and migrated client-side API calls to the batch application; reduced page load time from 10–15s to 2–3s and enabled daily data updates.
    • TI's customer email notifications were sent from two separate servers with inconsistent UI, risking spam classification as Google tightened sender requirements — developed a consolidated email alerts application, unifying delivery to a single server, standardizing the email UI across TI, and integrating one-click unsubscribe; eliminated the split-server issue and prevented TI emails from being marked as spam.
    • Business units had no mechanism to push near real-time product availability status updates, causing stale information across TI.COM — built backend capability for business units to update statuses, updated product detail pages to conditionally show information and suggested replacement parts, and re-configured internal search results based on product availability status.
    • TI needed to migrate customers from a legacy payment platform to a new Mendix/Boomi/SAP solution while maintaining trade compliance — built internal tooling to facilitate the migration, added customer profile review functionality for trade compliance, and maintained customer email notification integrations through RelationshipOne and Eloqua.
    JavaSpring BootTypeScriptStencilMendixBoomiSAP
  3. UF Human Neuromechanics Lab

    Research Assistant

    May 2022 – Dec 2022

    Gainesville, FL

    • Ankle prosthetic testing on amputees required real-time EEG signal streaming, but the existing plotter had 15ms latency — too slow for precise neuromechanical feedback; built a real-time plotter in C++ using SFML for graphics and Windows/Linux UDP sockets for data streaming, architecting the pipeline to minimize buffering and rendering overhead; reduced streaming latency from 15ms to 8ms, improving the accuracy and responsiveness of live signal visualization during amputee testing sessions.
    • Intact and residual lower-leg muscle activity in amputees needed to be tested across a distributed hardware setup, with a laptop handling high-level control logic and a Raspberry Pi handling low-level hardware interfacing — developed a control loop and controller in Python using sockets for bidirectional real-time data streaming between the two devices; enabled reliable, synchronized testing of muscle activity across both intact and residual limbs.
    C++PythonSFMLUDP SocketsRaspberry PiEEG
  4. Blue Sparq, Inc.

    Web Development Intern

    Jun 2021 – Aug 2021

    Cape Coral, FL

    • Redesigned company website from an existing HTML/CSS template; optimized performance and SEO ranking using Google Lighthouse.
    HTMLCSSSEO
AWS re:Invent 2025

Build Real-Time AI-Powered Customer Experiences with Amazon Bedrock and Twilio

Breakout Session71 AttendeesContact Center AILive Demo

Session Abstract

Poor customer experience costs businesses an estimated $3.7 trillion annually. In this session, we demonstrate how to architect a multi-channel agentic AI platform that normalizes voice, SMS, email, and web channels into a unified conversation stream — powered by Amazon Bedrock and Twilio ConversationRelay.

The reference architecture covers real-time AI conversation flows, autonomous agents for routine inquiries, and AI-augmented human handoff — all with persistent session state managed via DynamoDB and deployed as a single AWS CDK stack. Built with LangChain and TypeScript, the solution demonstrates production-grade patterns for routing, session management, knowledge base integration via RAG, and multi-modal channel normalization.

Channels

Voice · SMS · Email · Web

Unified stream via Twilio ConversationRelay + SendGrid

AI Layer

Amazon Bedrock

Claude via Converse API with Bedrock Knowledge Base (RAG)

Infrastructure

Single CDK Deploy

EC2 + ALB + DynamoDB + Amplify + Route53 + WAF

Amazon BedrockTwilio ConversationRelayLangChainAWS CDKDynamoDBRAGMulti-channel AIVoice AIContact Center
Open Source

Featured Project

An open-source sample published under aws-samples, demonstrating production-grade voice AI architecture on AWS.

aws-samples

sample-aws-bedrock-twilio-voice-ai

Multi-channel voice AI combining Amazon Bedrock with Twilio ConversationRelay. Build intelligent voice agents that understand natural language, maintain context, and integrate seamlessly with your existing communications stack — all deployed via AWS CDK.

Amazon BedrockTwilio ConversationRelayAWS CDKAWS LambdaTypeScriptVoice AI
Stars
Forks
View on GitHubApache-2.0 License
Get in touch

Let's Connect

Whether you're exploring cloud architecture, building with generative AI, or want to collaborate — I'd love to hear from you.

Send a Message

0/2000

Or email me directly at cameron@keenefl.com

Book Time

30-min Google Meet — pick a time that works for you.

Open to

  • Cloud architecture reviews
  • Generative AI & Bedrock consulting
  • Speaking / workshops
  • Open-source collaboration