AI Chatbot Developer

About the role Triangulum Insights runs an AI-powered analytics chatbot that gives clients conversational access to campaign research — POV methodology documents, creative diagnostics, and survey insights. The bot is a Python/Streamlit app using the OpenAI API, reading pre-staged data from AWS S3, deployed via Docker to a vendor-hosted portal (Zuar) and embedded in Tableau dashboards. You'll own the application end to end: code maintenance, prompt tuning, data pipeline coordination, and deployments. Responsibilities - Maintain and extend the chatbot application (Python, Streamlit, OpenAI API) - Tune and manage AI prompt templates that control the bot's behavior for each mode (POV library, creative diagnostics, qualitative verbatims) - Build and publish Docker images; coordinate deployments with the hosting vendor (staging first, then production) - Maintain the S3 data layer: shared POV library, per-client data folders, Q&A logs - Run the POV library rebuild process (R script) when methodology documents change; keep the POV-to-portal link mapping current - Maintain the client access-control gate (URL-parameter-based client validation — clients must never see each other's data) - Troubleshoot across the stack: app errors, S3/credential issues, portal embedding, country-detection logic - Coordinate with the Data Analytics team on the Tableau → S3 data pipeline (you don't run it, but you need to understand it) - Follow security practices: no credentials in code or git, secure secret handoffs, key rotation Required skills - Python — solid working proficiency; able to read and modify a ~3,500-line two-file codebase confidently - OpenAI API / LLM apps — prompt engineering, context management, structured (JSON) responses - AWS S3 — boto3, IAM credentials, bucket organization - Docker — building images, Docker Hub push/pull, docker-compose, debugging containers - Git/GitHub — comfortable with standard clone/branch/commit/push workflow Strongly preferred - Streamlit (or ability to pick it up fast — it's just Python) - R — enough to run and lightly modify existing scripts (the POV rebuild uses pdftools/jsonlite/aws.s3) - Tableau Cloud / REST API familiarity — understanding views, workbooks, PATs, and API rate limits - Experience working with a hosting vendor / managed deployment process (change requests rather than direct server access) Nice to have - WordPress/portal content APIs (the bot deep-links responses to portal posts) - Survey research / ad-effectiveness domain familiarity - Security hygiene mindset (secret scanning, credential rotation) What success looks like (first 30 days) - Runs the app locally (Python and Docker) and completes a full staging → production deployment through the vendor - Successfully rebuilds and publishes the POV library end to end - Makes a small prompt or UI change, tested on staging, shipped to production - Can explain the data flow (Tableau → pipeline → S3 → bot) and the access-control model unprompted