Stratify — AI market intelligence platform

• Built an ETL pipeline that ingests large Euromonitor, World Bank and World Population datasets and normalises them into one schema.
• Created analysis modules such as Market Size, Pricing Power and Growth, rendered as Recharts dashboards.
• Added an AI layer that turns queried data into trends, summaries and business recommendations.
• Gave users a single Okta login across dashboards, AI and exports.

Async PPT export

• Generated decks without blocking requests: a Celery task polls an external render service every 10s instead of holding a worker for the 3–5 min render.
• Made it crash-safe with acks_late and state split between Postgres and the render service, so a lost worker resumes polling with no double-submit.
• Uploaded finished decks to S3 and served them through presigned URLs.
• Cached decks by filter hash, so identical requests are served instantly.
• Used a pluggable builder per module, so adding a new deck means adding one file.
• Pushed live export status to the frontend over WebSockets (Django Channels).

Distributed tracing

• Traced Django, Celery, Redis, Postgres and OpenAI with OpenTelemetry and Jaeger.
• One end-to-end trace per user action — HTTP → Celery → LLM call → DB — with zero manual instrumentation.
• Recorded model and token usage on every OpenAI call to track LLM cost and latency.
• Toggled from config, with an OTel collector → Jaeger → Elasticsearch pipeline in production.

AI pipeline config control plane

• Imported the AI pipeline's S3 config into RDS as editable rows.
• Built an Angular admin console grouped by title and category, with enable/disable toggles.
• Wrote FastAPI update and delete APIs that edit the RDS copy, keeping changes as drafts until published.
• Added a global Publish that rebuilds the config from RDS and overwrites it in S3.
• Secured the console and APIs with Okta.