Case study
AI-Powered Quality Control Checker
Automated quality-control system that uses AI to evaluate customer calls and WhatsApp chat transcripts against SOP, KYC, response-time, product-explanation, and fraud-detection rules, then logs structured results into Google Sheets with email and Excel reporting.
- Project context
- Corporate Project
- Year
- 2026
- Categories
- AI Agent, Automation, Data Integration, Quality Assurance
- Source code
- Not publicly available
- Technology stack
- Python, Google Vertex AI, Gemini 2.5 Flash, Google Sheets API, Google Cloud Storage, PostgreSQL, SQLAlchemy, Pandas, Paramiko / SFTP, Pydub, OpenPyXL, SMTP Email
System design
A public-safe view of the components, integrations, and data flow described in this case study.
Scroll horizontally to inspect the full diagram, or open it full size.
Problem
Manual quality control for customer conversations was time-consuming, inconsistent, and difficult to scale across both voice calls and WhatsApp chats. The business needed a repeatable way to evaluate agent SOP compliance, response quality, payment-transfer safety, KYC completeness, and potential fraud signals without exposing customer data or relying entirely on manual review.
Role
Abdul designed and implemented the end-to-end automation, including the chat and call QC pipelines, AI prompts, structured response schemas, CRM and cloud-storage integrations, SFTP audio ingestion, Google Sheets reporting, email notifications, retry handling, logging, and scheduled execution setup.
Architecture
The system is a Python-based batch automation pipeline with two main entrypoints: - `chat_checker.py` for WhatsApp chat quality control - `call_checker.py` for call-recording quality control Core logic lives in `app.py`, split into: - `ChatQCChecker` - `CallQCChecker` The chat pipeline pulls paid case samples from a PostgreSQL CRM database, downloads chat exports from Google Cloud Storage, parses WhatsApp `.txt` or `.zip` transcripts, evaluates them with Gemini on Vertex AI, and writes QC results to Google Sheets. The call pipeline connects to SFTP recording servers, downloads and filters audio files, extracts metadata from filenames, optionally compresses audio, sends recordings to Gemini on Vertex AI, and writes structured call analysis into a Google Sheet report. Both pipelines include logging, idempotency checks, quota-aware retries, batch writes, email summaries, and Excel backup exports.
Implementation
Key implementation details include: - Google Vertex AI integration using Gemini 2.5 Flash with strict JSON response schemas. - Separate prompt files for call and chat evaluation rules. - WhatsApp transcript parser supporting multiple timestamp/export formats. - Smart chat truncation that preserves greetings, closing messages, payment keywords, and KYC-related sections. - PostgreSQL sampling query for selecting eligible paid chat cases. - Google Cloud Storage downloader for chat export files. - SFTP downloader for call recordings with support for multiple folder structures. - Agent-extension mapping through Google Sheets. - Call metadata extraction from recording filenames. - Audio duration and bitrate detection with optional 8kbps compression. - Batch Google Sheets writes with exponential backoff for quota/rate-limit errors. - Daily log rotation, Excel report generation, and HTML email status summaries.
Results
The project converted a manual QC workflow into a repeatable AI-assisted review system. It standardized how calls and chats are evaluated, centralized results in Google Sheets, added automated reporting, improved operational traceability, and introduced safeguards for fraud/payment-account detection and SOP compliance review.