Team Lead (AI & ML)

Qualification and Experience: Bachelor's or Master's degree in Computer Science, AI, Data Science, Software Engineering, or equivalent from a recognized university. Minimum 08 years of hands-on machine learning and AI development experience. Experience in government, law enforcement, or regulated-industry AI projects is strongly preferred. Note: Experience in private entities shall be accepted if such entity is registered with SECP, Registrar of Firms or any other Public Sector Regulatory Authority. Competencies & Personal Attributes: Demonstrated experience deploying and serving LLMs locally (not just API-based usage). Proven experience building and productionizing RAG pipelines. Hands-on experience with computer vision and/or audio/speech ML models. Deep intellectual curiosity — stays current with rapidly evolving LLM and AI research. Strong systems thinking — able to design end-to-end AI pipelines from data ingestion to prediction delivery. Resilient problem-solver in resource-constrained on-premises environments (no cloud fallback). Strong communicator — able to translate complex AI capabilities into business outcomes for executives. Commitment to data sovereignty, privacy, and government compliance requirements. Collaborative and able to work across software, infrastructure, and operations teams. Mandatory Technical Requirements LLM & Foundation Models: LLaMA, Mistral, Qwen, or equivalent — hands-on local deployment experience LLM Serving: Ollama, vLLM, llama.cpp, or TGI — inference optimization and quantization (GGUF/GPTQ/AWQ) RAG Frameworks: LangChain or LlamaIndex — end-to-end RAG pipeline implementation Vector Databases: Qdrant, ChromaDB, Weaviate, or Milvus — production-scale embedding storage and retrieval ML Frameworks: PyTorch, Hugging Face Transformers, PEFT (LoRA/QLoRA for fine-tuning) Computer Vision: YOLOv8/v10, OpenCV, ONNX Runtime — object detection and video analysis Audio/Speech: OpenAI Whisper / WhisperX, pyannote-audio — transcription and diarization MLOps: MLflow, DVC, or equivalent — experiment tracking, model registry, pipeline automation GPU Infrastructure: CUDA/cuDNN, NVIDIA GPU administration, VRAM optimization API Development: FastAPI (Python) or ASP.NET Core — for exposing AI services to existing software stacks Database: Microsoft SQL Server — for storing model outputs, metadata, and AI-derived analytics Languages: Python (primary), with working knowledge of C# for .NET integration Docker & Containerization: GPU-enabled Docker (NVIDIA Container Toolkit), docker-compose AI Productivity Tools: Claude, GitHub Copilot, or equivalent — mandatory daily usage

Areas of Responsibilities: Local LLM Infrastructure & Hardware Design and procure GPU hardware specifications for the PERA on-premises AI data center (NVIDIA A100/H100 or equivalent). Install, configure, and maintain the full local AI stack — CUDA, cuDNN, drivers, and inference runtimes (Ollama, vLLM, llama.cpp). Deploy and manage open-source foundation models (LLaMA 3, Mistral, Qwen, or equivalent) entirely on local hardware. Implement model quantization (GGUF, GPTQ, AWQ) for efficient inference within available GPU VRAM constraints. Ensure zero data egress — all model inference, training data, and embeddings remain on PERA servers. RAG System & Knowledge Base Design and implement a production-grade Retrieval-Augmented Generation (RAG) pipeline using LangChain or LlamaIndex. Build a local vector database (Qdrant, ChromaDB, or Weaviate) to store embeddings of PERA documents, regulations, and operational data. Develop document ingestion pipelines supporting PDF, Word, structured DB data, and scanned documents (via OCR). Implement semantic search, hybrid search (dense + sparse), and re-ranking for accurate knowledge retrieval. Create a self-improving feedback loop where user interactions are used to refine retrieval and generation quality. Model Training, Fine-Tuning & Continuous Learning Fine-tune foundation models on PERA-specific datasets using LoRA/QLoRA to adapt them to domain language (legal, enforcement, regulatory). Design data pipelines to collect, clean, label, and version-control training datasets from operational software logs and databases. Implement continuous model evaluation, A/B testing, and performance benchmarking (ROUGE, BLEU, F1, custom domain metrics). Maintain a Model Registry for versioning, rollback, and lifecycle management of all deployed models. AI APIs & Software Integration Develop and expose AI capabilities as secure RESTful APIs (FastAPI / .NET Core) for consumption by FRMS, PERA360, HRMIS, and other platforms. APIs to deliver: natural language Q&A, document summarization, intelligent search, prediction outputs, and anomaly detection. Implement forecasting and analytics models (time-series, regression) for enforcement trends, inspection planning, and resource allocation. Ensure API versioning, authentication, rate limiting, and API documentation (Swagger/OpenAPI). Computer Vision & Object Detection Deploy and fine-tune object detection models (YOLOv8/v10, RT-DETR) on PERA's GPU infrastructure for video-based surveillance. Build real-time and batch video analysis pipelines for detection of violations, unauthorized activities, and entity recognition. Integrate computer vision outputs with FRMS enforcement workflows via API hooks. Maintain model accuracy through periodic retraining on new annotated video datasets. Audio Intelligence — Transcription & Keyword Detection Deploy local speech-to-text models (Whisper / WhisperX) supporting Urdu, Punjabi, and English for body-camera and call recordings. Build keyword spotting and flagging pipelines to detect specific phrases, commands, or compliance-relevant terms in audio streams. Develop speaker diarization capabilities to identify and separate multiple speakers in recordings. Integrate audio intelligence outputs into PERA360 officer accountability workflows. Data Management & Compliance Design SQL Server schemas and data warehouses to store AI inference results, model outputs, and training metadata. Enforce strict data governance — classification, access control, audit logging, and retention policies. Ensure all AI pipelines comply with Government of Punjab data residency and cybersecurity requirements. Produce regular AI system performance reports for leadership and compliance documentation. Team Leadership Lead and grow the AI/ML team; define roles, onboarding plans, and technical growth paths. Establish MLOps practices: experiment tracking (MLflow), model versioning, automated retraining pipelines. Collaborate with backend, frontend, and DevOps teams to embed AI features into production software. Produce technical documentation, architecture diagrams, and research summaries for stakeholder communication.