Policy Assistant
Uploads document content, extracts text server-side, combines it with user prompts, and uses Gemini to answer questions precisely.
A production AI platform that combines Gemini and AssemblyAI for document Q&A, knowledge analysis, meeting transcription, and survey intelligence inside one consistent user workflow.
The value was not just API usage. It was the workflow design, file handling, persistence, and user-facing control around those AI capabilities.
Uploads document content, extracts text server-side, combines it with user prompts, and uses Gemini to answer questions precisely.
Combines multiple uploaded files into one AI analysis pipeline with custom prompts and persisted output histories.
Handles notes directly or transcribes audio through AssemblyAI before sending results to Gemini for structured outputs.
Builds surveys, creates public share links, collects responses without login, and turns feedback into analyzed insights.
Text intelligence for Q&A, synthesis, and structured outputs.
Audio transcription and speaker-aware meeting processing.
Workflow orchestration, file extraction, and session control.
Persisted histories across tools, files, surveys, and results.
Transactional account and workflow notifications.
Hardened session handling and ownership checks.
Audio transcription does not finish instantly and can outlive naive request lifecycles.
Built a polling loop with controlled retries and clear status exits before handing results to Gemini.
DOCX, TXT, CSV, and audio all need different extraction paths.
Implemented format-specific handlers instead of forcing one generic parsing flow onto everything.
Survey submission should be public while management remains protected.
Used share tokens for public access and ownership checks for all management surfaces.
Default session behavior is not enough for a production-facing application.
Strengthened session handling with stricter lifecycle control and safer storage assumptions.
Shows the ability to chain AI providers into real user workflows.
Covers creation, collection, and analysis inside one product.
Purpose-built handling for different input types instead of shallow API wrapping.
Quadri AI reflects the kind of work I like: turning raw AI capabilities into stable, repeatable workflows that people can actually use.
I focus on the orchestration layer, persistence, access control, and interaction design that make AI features feel like a real product.
Open to product-focused backend and full-stack roles involving real workflow complexity.