AIdeas: REGAIN - Your Professional Edge.

Social Good

Goldman Sachs has estimated that AI could potentially automate the equivalent of about 300 million full‑time jobs globally. Every year, 200,000 service members transition out of the military. Two seemingly distinct worlds share one identical challenge. The hardest part of career transition is not finding job listings but rather the crisis of professional identity. Who am I now? What am I qualified for? How do I prove what I can do?

REGAIN answers those questions with action.

REGAIN is an AI-powered career reskilling platform that runs personalized campaigns for veterans, AI-displaced workers, and career pivoters. It rebuilds the professional identity that makes someone ready to pursue a new career and discover newfound purpose.

The platform assigns daily skill-building missions calibrated to the user's background, target role, and current "campaign phase". Every completed mission generates timestamped evidence. This serves as documented proof of demonstrated capability stored in the user's Evidence Vault. An adaptive coaching agent monitors progress, adjusts difficulty, flags avoidance patterns, and provides market-aware guidance. A market intelligence pipeline tracks real-time demand for the user's target role, ensuring missions align with what employers actually need.

The result is a professional who has the documented receipts to prove readiness. REGAIN even auto-generates an agent-friendly markdown resume with structured frontmatter, so AI recruiting agents can parse, score, and surface the user's profile without human intervention.

I built this tool because I lived this challenge.

As a former Army Green Beret who transitioned out of the military, I know what it feels like to have a decade of proven capability and no way to translate it. REGAIN is the solution to this problem.

Career transition presents people with a double-pronged problem: a confidence crisis and a logistics problem.

Every existing tool (job boards, resume builders, networking platforms, etc.) assumes you already know your value and just need to broadcast it. They optimize for distribution of talent, but for someone who just left the military after 12 years, or for a senior marketing manager whose role was eliminated by AI, or for a factory worker whose plant closed, the talent distribution problem is secondary. The primary problem is one that often presents itself with this thought: "I don't know what I'm worth anymore."

REGAIN addresses this gap directly. Through structured daily missions, each designed to surface and document a specific skill, users build an evidence-based portfolio of demonstrated capability.

This matters at scale because the problem is growing. AI displacement is happening now across content creation, customer service, data analysis, and administrative roles. The workforce transition challenge of the next decade will undoubtedly dwarf anything current job boards can handle. Given this, people need a system that meets them in the identity crisis and walks them out of it with documented proof of who they've become and the value they can add to an evolving workforce.

Ultimately, the social impact REGAIN makes is a measurable one: missions completed, evidence logged, skills demonstrated, and market alignment scores improved.

REGAIN was built using spec-driven development in Kiro, with infrastructure deployed entirely on AWS.

The application follows a serverless, event-driven architecture. A React frontend hosted on AWS Amplify serves the user interface at regain.altivum.ai . Amazon Cognito handles authentication with email-based sign-up and SRP auth flow. AWS Lambda functions process all business logic including onboarding, mission generation, evidence logging, coaching, and market intelligence, with API Gateway exposing secure REST endpoints and two WebSocket APIs handling real-time interactions. Amazon DynamoDB stores user profiles, campaigns, mission history, evidence vault entries, market data, voice session records, and WebSocket connection state across seven purpose-built tables, each with least-privilege IAM access per Lambda.

REGAIN's coaching agent is built with the Strands Agents SDK and powered by Amazon Nova Lite. The agent operates with 13 tools that read and write directly to DynamoDB, giving it full context on the user's profile, campaign status, mission history, evidence vault, and market alignment. It autonomously generates missions, logs evidence, adjusts difficulty, evaluates phase gate criteria, and provides market-informed career guidance. Text coaching streams in real time over WebSocket, with tool execution visibility so users can see the agent thinking and acting on their behalf.

Voice capabilities use Amazon Nova Sonic for bidirectional speech-to-speech streaming across two distinct experiences. Voice onboarding lets users complete their entire intake by talking to their AI coach instead of filling out forms. Voice practice provides structured mock interviews and mission discussions. On disconnect, the system generates a detailed assessment via Nova Lite, scoring articulation, confidence, and content quality, then stores the transcript and assessment to S3.

The mission engine runs a 7-stage pipeline: data retrieval, skill gap analysis, difficulty state evaluation, template instantiation, difficulty filtering, duplicate exclusion, and score-based ranking. It produces a primary mission and two alternates from five categories: reflection, skill building, portfolio development, networking, and market research. Phase gates evaluate after each completion: 10 missions advance from Foundation to Expansion, 20 to Launch, 30 to completion. Difficulty adapts per-category on a 1-5 scale based on consecutive completion patterns.

The market intelligence pipeline runs on EventBridge-scheduled Lambda functions that ingest data from O*NET and USAJobs, scoring role demand, tracking growth trends, and identifying in-demand skills. This data feeds the coaching agent's mission generation, ensuring every mission aligns with real market needs.

AgentCore Gateway routes tool invocations through a centralized MCP endpoint with Cognito JWT authorization and 12 registered tool schemas. The infrastructure supports Cedar policy enforcement for fine-grained access control through features like user data isolation, evidence write scope, mission generation rate limits, profile update restrictions, and market data read-only policies. AgentCore Observability instruments every coaching session with structured traces exported to CloudWatch, capturing session duration, tool invocation latency, and span-level detail for operational dashboards.

Kiro's spec-driven workflow generated structured requirements, technical designs, and sequenced implementation tasks across five specifications: Foundation Infrastructure, Coaching Agent, Mission Engine, Market Intelligence, and AgentCore Platform Integration. Each spec followed a requirements → design → tasks pipeline, with design reviews catching architectural issues (like Cedar policy evaluation dependencies on database queries) before implementation began.

The frontend was built in parallel using Claude Code. These pages include Onboarding, Dashboard, Missions, Evidence, Coaching, Voice Practice, Voice Session Detail, Profile, Resume, and O*NET Career Explorer. A design system with self-hosted General Sans and JetBrains Mono fonts, CSS-only animations, and a WebGL 3D audio visualizer orb for voice sessions gives the interface a polished, production-grade feel.

Adding Nova 2 Sonic voice capabilities transformed REGAIN from the status quo of AI tools into a coach you talk to and truly interact with. Speaking about your career transition out loud, hearing yourself articulate what you've done and where you're going, is a fundamentally different experience than typing into form fields. For the target audience, removing standard forms helps them feel truly heard. The voice practice feature takes this further: users rehearse articulating their skills in mock interview settings, building verbal confidence alongside documented evidence.

Spec-driven development can prevent expensive mistakes but articulating intent matters.

Kiro's requirements → design → tasks workflow caught architectural issues early. A design review for the AgentCore integration identified that Cedar policies were attempting database queries during policy evaluation which is something that would have surfaced as a production latency problem weeks later. Specs cover what gets built but not necessarily what stays running. An environment management gap caused a deployed User Pool to go missing, breaking the entire auth flow. The lesson I learned: spec-driven development needs to extend into operational concerns, not just code generation.

The agent-friendly resume with structured YAML frontmatter is a small feature with outsized implications. As AI recruiting agents become common, having a machine-readable professional profile that updates automatically with evidence-backed accomplishments positions REGAIN users ahead of everyone still uploading PDFs to job boards. Building for the agent-mediated future is the practical approach to the future of job hunting.