Palak Singla

AI Engineer with nearly a decade of hands-on experience training and shipping production ML systems across computer vision and large language models. Personally owned the full ML lifecycle - data engines, training runs on multi-GPU clusters, evaluation design, deployment, monitoring, and iteration - for detection/segmentation systems (Scaled-YOLO, Mask R-CNN, Wide ResNet) and fine-tuned open-weight LLMs (PEFT/LoRA, quantization, TGI/vLLM serving).

Operates as a forward deployed engineer: embedded with clients, diagnosing real-world model failures (lighting, occlusion, camera angle variance) and closing the loop with hard-example mining and retraining. Repeatedly the first or solo ML hire, comfortable owning AI end-to-end in early-stage, high-ownership environments.

Senior Machine Learning Engineer II at Findem AI (2022-08 – 2026-07)

• Core technical lead through Findem's growth to its Series C raise - designing and deploying the scalable AI architectures behind the platform, processing 21M+ complex records with high accuracy in production.
• Engineered enterprise data pipelines centralizing data from Salesforce, Zendesk, Stripe, and 100+ MongoDB collections.
• Developed and integrated a Generative AI Copilot into the client-facing platform, translating complex search and filtering workflows into intuitive, natural-language solutions for users: designed the multi-agent AI system powering the copilot, with Pinecone as the vector database for semantic retrieval.
• Leveraged Snowflake AI to execute advanced data aggregations and predictive performance analysis, directly impacting business forecasting.

Senior Data Scientist - Computer Vision (sole owner of CV model stack) at ClaimGenius Pvt Ltd (2020-09 – 2022-08)

• Owned CV models end-to-end in production for AI-based vehicle damage assessment: personally trained Scratch Detection, SUV Parts, and Minivan Parts instance-segmentation models (Mask R-CNN) and detection models (Scaled-YOLO) - making all training and data decisions, not fine-tuning through APIs. Initial models were trained on 8,000 client-supplied claim images, then continuously fine-tuned on new images received from our major client, Unit Insurance Company (US).
• Built the data engine behind the models: curated training datasets from real customer claim photos and defined the labeling strategy - deciding which vehicle parts and damage types to annotate - starting with LabelMe and then building a custom in-house annotation tool to scale labeling; designed evaluation/test sets that mirrored real-world claim conditions so offline metrics predicted production accuracy.
• Diagnosed real-world failures and iterated: the biggest production challenges were shadows falling across car bodies and overexposure from direct sunlight, along with glare on reflective surfaces, partial/occluded views, and uncontrolled customer camera angles; mined these hard examples back into training data and tuned post-processing, improving segmentation accuracy by 12-15%.
• Owned the full loop of data preprocessing, augmentation, hyperparameter tuning, training runs, evaluation, deployment, and post-deployment monitoring; retrained and shipped improved model versions based on production feedback.
• Built and productionized a TensorFlow classification model determining vehicle insurance eligibility from visual damage analysis. Awarded "Star Performer of the Year"; met all quarterly targets ahead of schedule.

Artificial Intelligence Engineer - Computer Vision at Zentrum Technologies Pvt Ltd (2018-08 – 2020-06)

• Implemented the Wide ResNet research paper to build gender and age recognition models (with face detection) in TensorFlow, deployed for real-time inference on live camera feeds in commercial offices and warehouses: uncontrolled environments with variable lighting, camera placement, and occlusion.
• Built the training dataset of 250,000 (2.5 lakh) annotated images - combining open-source datasets with a substantial portion annotated in-house; owned dataset preparation, model training, deployment, and real-time inference performance, iterating on failures observed at customer sites.
• Led the technical team on a secure Government AI initiative, building the HPC environment on AWS: TensorFlow GANs generating human-like password candidates, plus genetic algorithms and simulated annealing for cryptanalysis.

Bachelor of Technology in Computer Science – Lovely Professional University

Nanodegree in Deep Learning – Udacity

Nanodegree in AI in Trading – Udacity