Rohit Jain | Senior AI Engineer

01About

Senior Artificial Intelligence Engineer with a master's degree in AI from the National University of Singapore and 7 years of experience building production-grade Generative AI, Computer Vision, NLP, and multimodal solutions across the Finance, IT, Agritech, Construction, and Automotive industries.

Skilled in architecting, developing, and deploying end-to-end ML systems, from research and model training to containerized microservices on cloud infrastructure.

02Experience

Senior Data Scientist

Meinhardt Group

2025 - Present Singapore

MultimodalAgentic RAGGraph RAGRAGASLangGraphMLOpsAzure

Developed a Multimodal Agentic RAG system for Q&A over technical documents and engineering drawings, iterating from vector RAG to Graph RAG to resolve accuracy degradation on complex, relational queries, improving multi-hop query accuracy by 35%+ over baseline vector retrieval.
Designed a Graph RAG pipeline that extracts entities, relationships, and key claims from engineering documents into a knowledge graph, enabling multi-hop reasoning across interconnected concepts and reducing hallucination by 40%+ by overcoming the structural context loss of chunk-based vector retrieval.
Implemented hierarchical, structure-aware chunking to preserve tables and cross-references in engineering documents, replacing fixed-token splitting and cutting context-fragmentation retrieval errors by 30%+.
Benchmarked RAG pipeline quality using RAGAS (faithfulness, context precision, context recall, answer relevancy), achieving 95%+ accuracy, 99%+ consistency, and <5% hallucination rate across production evaluation sets.
Engineered a scalable microservices architecture, containerizing ingestion, partitioning, and query pipelines with Docker into 10+ independently deployable services.
Deployed and orchestrated the application on a self-managed Kubernetes cluster, ensuring 99.9%+ uptime and efficient auto-scaling under variable load (100+ concurrent requests).

Senior AI Data Scientist

Ailytics AI

2024 - 2025 Singapore

PyTorchOpenCVML PipelineVision LLMTrackingPose EstimationMLOps

Built an AI video-analytics solution for worksite safety and productivity, implementing state-of-the-art object detection, multi-object tracking, and human/facial pose-estimation models with 90%+ precision and recall.
Designed and deployed a Multimodal Vision LLM (image-text-to-text) that flags safety violations in non-rule-based scenarios by analyzing images against user-configurable textual prompts.
Accelerated inference through model optimization and quantization to ONNX, TensorRT, and NCNN formats.
Created an end-to-end ML pipeline in ClearML covering dataset versioning, model training, evaluation, and benchmarking.

Artificial Intelligence Specialist

Vulcan AI Agritech

2022 - 2024 Singapore

SmartGrader

DetectionTrackingSegmentationClassificationTensorFlowAzureGitHub Actions

Designed and implemented edge-deployed AI for automated fruit counting and grading using computer vision.
Built object detection, tracking, segmentation, and classification models across agritech solutions, achieving 99%+ counting accuracy and 95%+ grading accuracy while significantly reducing manual labor.
Optimized inference through model quantization, cutting model size by 40% and improving inference speed by 60%.
Developed a CI/CD pipeline using Docker, Azure Container Registry, and GitHub Actions.

Chat with Fruit Doctor

Large Language ModelsGenerative AIPEFTLangChainHuggingFace

Implemented an LLM-based chat platform enabling farmers and stakeholders to query information on fruit health, growth, and nutrition.
Fine-tuned a 7B-parameter LLM on a domain-specific dataset, achieving high accuracy on agricultural Q&A.
Curated a 10K-sample fine-tuning dataset compiled from domain experts and government guidelines.
Improved answer quality by incorporating Retrieval-Augmented Generation (RAG) over 100+ research papers.

Super Resolution of Satellite Images

GANGenerative AIRemote SensingPythonPyTorch

Designed a patented Super-Resolution GAN that generates high-resolution (HR) crop imagery from low-resolution satellite images, achieving 5× upscaling.
Reduced annual HR-imagery acquisition cost by 12× and improved crop health and nutrient prediction performance by 50% using the generated HR images.
Improved output quality by designing a custom loss function with perceptual (feature) loss, optimizing for visual similarity over pixel-space metrics.

AI Engineer (part-time, during master's)

Panasonic R&D Center Singapore

Jan - Jun 2022 Singapore

Autonomous Vehicle Vision

Neural NetworksOpenCVGANPyTorchVideo Analytics

Built a deep learning model to detect street objects occluded by raindrops, enabling safety-critical toggling of autonomous vehicle (AV) self-driving mode.
Applied GANs for image de-raining alongside OpenCV image-processing techniques to minimize raindrop occlusion across varying sizes, shapes, and intensities.
Implemented a Siamese neural network for one-shot verification of predicted objects at inference.

Topic Transition System

Speech ProcessingNLPHuggingFaceAWS

Built a topic-transition and sentiment-detection system for analyzing conversations in a meeting-room environment.
Developed a state-of-the-art Wav2Vec2-based Automatic Speech Recognizer (ASR) dedicated to the Singaporean English accent.
Achieved a 5.9% word error rate (WER) on the NSC dataset, outperforming the leaderboard Wav2Vec2 baseline.
Applied NLP techniques to ASR-transcribed text to design the topic-transition detection model.

Data Scientist Intern

Staple AI

Apr - Jul 2021 Singapore

OpenCVFeature DetectionNLPFlaskDocker

Built image-processing pipelines for document image correction using OpenCV and traditional computer vision techniques.
Designed deep learning models to identify key features within documents using CNNs such as Faster-RCNN and Mask-RCNN.
Enhanced data-extraction algorithms for financial documents; applied NLP techniques to compare extracted data against ground truth and visualized results with Plotly.
Developed REST APIs and deployed scalable, dockerized Flask applications on AWS.
Managed independent projects with direct client engagement across multinational banks.

Data Scientist

Hewlett Packard Enterprise R&D

2018 - 2021 Bangalore, India

Linear RegressionDecision TreesBoostingMLOps

Designed, modelled, and maintained an AI-powered prediction engine that proactively detects and auto-resolves infrastructure issues spanning availability, performance, security, and capacity.
Built and automated an end-to-end ML pipeline on HPE Ezmeral for data ingestion, cleaning, feature engineering, model training, and deployment.
Proposed new data-collection points and developed interfaces to ingest sensor data.

03Skills

Areas of Expertise

Generative AILarge Language Models (LLMs)Retrieval-Augmented Generation (RAG)AI AgentsPrompt EngineeringFine-tuningEmbeddingsMachine LearningDeep LearningComputer VisionNatural Language ProcessingMLOps / LLMOps

Languages & Databases

PythonPostgreSQLMongoDBVector Databases

ML Libraries

PyTorchTensorFlowTransformersHuggingFaceSentenceTransformersOpenCVspaCyNLTKscikit-learnPandas

LLM Frameworks

LangChainLangGraphLlamaIndexOpenAI APIRAG Pipelines

Tools & Cloud

DockerKubernetesMLflowClearMLFlaskFastAPIGitGitHubAWSAzure

04Projects

Video Visual Relationship: Video Captioning

Designed a seq2seq architecture to identify visual object-relationship pairs in video, achieving 75% top-5 accuracy.
Extracted and sequenced video-frame features using InceptionV4 and ResNet128 CNN models.
Modelled a bidirectional, multi-stacked RNN decoder to generate captions; improved overall accuracy via model ensembling.

CNNRNNInceptionV4ResNet128

Question Answering on Stanford SQuAD v2.0

Built a transformer-based system to answer questions from Wikipedia reading comprehensions.
Explored architectures including BERT, ALBERT, ELECTRA, RoBERTa, and XLNet as base models for the QA system.
Achieved a top-35 global leaderboard rank with an F1 score of 87.18 using a custom fine-tuned RoBERTa-large.

BERTRoBERTaTransformersNLP

Live Face Mask Detection

Built a 98%-accurate real-time face-mask detector that handles masks of any pattern and colour.
Applied transfer learning with a MobileNet v2 model on extracted faces to classify mask vs. no-mask.

MobileNet v2Transfer LearningOpenCVComputer Vision

05Education

National University of Singapore

Master in Computer Science, Artificial Intelligence specialization

2021 - 2022 CAP 4.45 / 5 · Singapore

Manipal University

Bachelor of Technology in Electrical & Computer Engineering

2014 - 2018 CGPA 9.36 / 10 · Karnataka, India

06Certifications & Achievements

Certifications

Achievements

Received the HPE Employee Recognition Award three times for innovation and impact delivered through work.
Awarded the Director's Medal for Academic Excellence, finishing in the top 2% of a class of 190 students.
Finished in the top 1% of all participants in the 10^th International Mathematical Olympiad (IMO).
Won 2^nd prize in an eye-controlled robotics event at the Indian Institute of Technology (IIT) Madras.