Description :

The engineer is expected to support GTB in executing an independent, AI-assisted content safety review. The support would be required for building and deploying an independent detection engine to identify NSFW and inappropriate content across text chat, voice, and video interactions.

Work Requirements :

The engineer will support in building, calibrating, and deploying GT's independent content review engine. The engagement would be using an LLM (frontier models) for classification. The engineer will be working with AI tooling extensively.

Specifically :

- Reviewing the behaviour category taxonomy and translating it into a detection schema

- Building a two-layer detection pipeline : regex and rule-based layer for pattern matching, followed by an LLM classification layer using LLM for contextual detection

- Designing and integrating a voice tone analysis module for aggression, stress, and hostility detection from audio samples

- Constructing and documenting the evaluation set labelling schema : the ground truth framework that analysts will use for manual annotation

- Running calibration sessions with the analyst team to ensure annotation consistency

- Validating engine output : precision, recall, false positive rate, threshold tuning

- Orchestrating the full pipeline run against the sample data on client infrastructure

- Preparing the benchmarking output between GT's engine and the client's existing detection system

Must-Have Skills :

- Strong Python pipeline development, API integration, data processing

- Hands-on experience with LLM APIs (Gemini, OpenAI, or equivalent) prompt engineering, classification workflows, output parsing

- Familiarity with coding agents and AI-assisted development workflows

- Experience with text classification and NLP tasks

- Basic audio processing feature extraction using libraries such as librosa or equivalent; ability to integrate pre-trained audio models

- Ability to work within hardware constraints (CPU-only, 8GB RAM)?

Good to Have :

- Familiarity with multilingual content the sample will contain multiple Indian languages alongside English

- Experience building evaluation sets or ground truth datasets for ML systems

What the Engineer Does Not Need to Handle :

- Cloud infrastructure or DevOps

- Frontend or UI development

- Manual annotation