Building production-ready serverless applications with modern AWS services and infrastructure as code
Building scalable, cost-optimized serverless solutions with modern frameworks
Experimenting with LLMs, model context protocols, and AI-powered applications
Fully serverless RAG document intelligence platform on AWS. Ingests PDFs, DOCX, audio (AWS Transcribe, multi-language), and live URLs. Hybrid OCR: Claude Vision for short scans, Textract for longer ones. Titan Embeddings + pgvector power semantic search and source-cited, streaming chat (Claude 3.5 Sonnet) with scoped multi-document comparison and persistent session history. Admin panel covers Cognito user management, per-model token cost analytics, and RDS start/stop for cost control. 22 Lambda functions in a private VPC, full SAM IaC. PWA-ready.
↗ launchServerless ticket-routing engine driven by a configurable, priority-based rules table. Tickets fan out through EventBridge to per-team SQS queues (engineering, billing, product, escalation) with dead-letter queues, while Step Functions handle SLA monitoring and automatic escalation. Includes a web dashboard for managing tickets and routing rules, with Cognito role/team-based access over a REST API.
↗ launchServerless zero-knowledge credentials vault on AWS. Client-side Web Crypto API encryption with Argon2i key derivation — the server stores only ciphertext and never sees a plaintext key. Manages local vault entries and live AWS credentials: full CRUD against Secrets Manager secrets and SSM Parameter Store parameters, with bidirectional sync. Includes a cryptographically secure password generator, .env file importer, auto-lock, and password history enforcement.
↗ launchIncident management platform for engineering teams - a lightweight PagerDuty hybrid. Per-incident and cross-incident RAG chat, AI living timeline summaries, dual-format AI postmortems (Executive + Developer) emailed to stakeholders, action items with email assignments, role-based access control, admin audit log, MTTR analytics, and PWA support.
↗ launchLocal-first codebase intelligence tool. Scans a workspace and reports per-project languages/LOC, dependencies, and known vulnerabilities (OSV.dev) with a deterministic health score. Adds local-LLM insights (Ollama): project health narratives, copy-paste remediation commands, and source-grounded AI migration reviews for major-version upgrades. npm, Python, Maven, and Go ecosystems.
Serverless accounts receivable platform with a multi-seller, multi-buyer company model. KPI dashboard with critical-account flagging, 6-month collections trend charts, top outstanding balances, and per-seller breakdowns. Full payment lifecycle with CSV import/export for balances and companies, invoice generation, and EN/ES localization. Weekly rich-HTML email digest (EventBridge cron → SES) with KPI snapshot; self-service SNS subscription from the app. Role-based access (admin/viewer) via Cognito groups with GSI-backed seller-scoped data isolation.
↗ launchAutonomous market intelligence dashboard tracking tech salary trends, job demand signals, and remote work patterns for your target roles. Weekly data collection via scheduled agent, AI-generated market summaries, resume analysis and scoring, cover letter generation, JD matching, and RAG-powered career chat.
↗ launchMulti-cert AI study platform. Generates quizzes from a curated question bank, an LLM grounded in your uploaded study materials (RAG via pgvector), or a specific document — single-choice and multiple-response, with a timed exam-simulation mode and mistake-targeted practice that drills your own wrong answers. Spaced-repetition flashcards (SM-2) with a per-card AI tutor, readiness scoring, score-trend charts, AI weekly study plans, mistake digest, RAG-enhanced concept explainer, weekly progress email digests, curated intel feed, and full PDF export. Self-service cert onboarding via Cert Forge.
↗ launchVoice-first personal assistant running fully local — no cloud LLM. Wake-word or push-to-talk loop: openWakeWord → Deepgram speech-to-text → a LangGraph ReAct orchestrator that routes to specialized sub-agents (email, calendar, code, general — more on the way) over Gmail / Calendar / Git tools → Deepgram text-to-speech. ChromaDB-backed semantic memory, runtime model hot-swap across Ollama models, and a real-time React presence UI over a FastAPI WebSocket — including a Three.js per-agent orb that visualizes which agent is thinking. Private by design: conversations and tools stay on-device.
Growing Chloe into a fully private, self-hosted assistant — additional specialized sub-agents, richer tool integrations, and smarter on-device memory, all on local hardware with no cloud LLM.
Interactive demos and documentation for key projects from BJPlayground
Sharing lessons learned from AWS optimization, serverless architecture, and AI integration