AI/ML Engineer

Apply through the official VOPA careers portal: https://careers.vopa.in/jobs/ai-ml-engineer-38Vowels of the People Association (VOPA) is hiring an AI/ML Engineer in Kothrud, Pune, for a full-time, in-office role. Freshers with strong practical AI/ML skills and candidates with 0–3 years of experience are welcome. VOPA builds AI-powered education technology, including its NIPUN Maharashtra platform, which supports foundational literacy and numeracy assessments across more than 70,000 schools and serves 5 million students. The role focuses on fine-tuning open-source models, building machine learning pipelines, and deploying production-ready AI systems for handwritten answer evaluation, Marathi speech recognition, computer vision for Indic scripts, and efficient AI inference. This is a hands-on engineering and research role, not a data annotation or prompt engineering position.ResponsibilitiesFine-tune open-source language and vision models using LoRA, QLoRA, and other parameter-efficient fine-tuning techniques.Work with Hugging Face Transformers, Unsloth, and PEFT libraries.Optimize GPU training performance and resource utilization.Experiment with model architectures for educational applications.Develop models for Marathi handwriting recognition and answer evaluation.Build speech recognition systems for Marathi and other Indic languages.Design AI-based rubric scoring and evaluation mechanisms.Handle noisy, inconsistent, and low-resource datasets.Deploy fine-tuned models to AWS infrastructure, including ECS/Fargate and GPU inference endpoints.Optimize inference latency, throughput, performance, and infrastructure cost.Explore quantization, batching, ONNX, TensorRT, and on-device AI deployment.Build end-to-end ML pipelines for preprocessing, augmentation, training, evaluation, and serving.Track metrics such as Word Error Rate (WER), Character Error Rate (CER), and Devanagari-aware benchmarks.Conduct structured error analysis and improve model reliability through experimentation.Review AI/ML research papers and translate relevant ideas into practical implementations.Collaborate with backend engineers to integrate models into FastAPI services.Contribute to technical decisions involving model architecture, training, and deployment.RequirementsStrong understanding of Transformer architectures, attention mechanisms, embeddings, and positional encoding.Knowledge of encoder, decoder, and encoder-decoder architectures.Hands-on experience with LoRA/QLoRA fine-tuning using Hugging Face PEFT or Unsloth.Understanding of quantization techniques such as NF4, GPTQ, and AWQ.Familiarity with learning rate scheduling, gradient accumulation, and mixed-precision training.Ability to diagnose training instability, loss spikes, and gradient-related issues.Good programming skills in Python or Rust.Proficiency in PyTorch, including autograd, tensor operations, and custom loss functions.Experience with computer vision techniques such as CNNs, image preprocessing, and augmentation, and/or speech AI techniques such as CTC, attention-based models, and audio feature extraction.Understanding of dataset preparation, labeling strategies, and train/validation/test splits.Ability to write clean, modular, production-oriented code.Experience deploying ML models using Docker and cloud infrastructure.Familiarity with GPU-based inference environments.Understanding of inference optimization, batching, ONNX, TensorRT, and KV caching.Awareness of model size, latency, and runtime constraints for edge devices.Ability to balance inference cost, latency, and throughput.