Projects
Below are the experiments I am currently building or planning. Each one is intentionally scoped so I can deliver a demo-worthy slice in a few weekends.
GenAI News Copilot · 2025
Streamlit front-end that answers questions about curated news using retrieval-augmented generation, guardrailed APIs, and an AWS-backed document store.
Architecture sketch:
- Python crawler (feedparser + BeautifulSoup) collects articles, normalizes text, and stores documents in S3 plus metadata in DynamoDB/OpenSearch.
- Indexer lambda chunks articles, generates embeddings (AWS Bedrock or local model), and registers them in a vector store.
- Guardrails layer wraps LLM calls, filters responses, and ensures every answer cites sources.
- Streamlit app handles auth, issues semantic search queries, calls the guardrailed API, and displays answers with citations.
Data & AI Platform Starter · 2024
Ship an infrastructure-as-code template that spins up ingestion (Kafka + Debezium), transformation (dbt + Dagster), and serving (DuckDB/Delta tables) so analytics teams can bootstrap trustworthy metrics quickly.
Architecture sketch:
- Wire Kafka, Debezium, Postgres OLTP, and DuckDB via docker-compose.
- Add Dagster jobs to move CDC topics into an object store and trigger dbt models.
- Publish a sample dashboard (Metabase or lightweight Streamlit) powered by the curated metrics.
React CRM Playground · 2022-2023
A polished React SPA (likely Next.js + Chakra UI) that demonstrates contact management, a deal kanban, and activity timelines using mocked API services.
Architecture sketch:
- Define personas and sample datasets for contacts, opportunities, and tasks.
- Build reusable UI primitives (avatar stacks, kanban columns, activity feed) with responsive design.
- Add MSW scenarios for optimistic updates, errors, and latency to simulate real-world API behavior.
Have something interesting I should explore or collaborate on? Reach out and let’s chat.