CV
Education
Dec 2018 – Nov 2022
Ph.D., Computer Science (Computer Vision)
Friedrich-Alexander-Universität Erlangen-Nürnberg
- Chair of Pattern Recognition. Dissertation on scene understanding in digital humanities.
- Cumulative GPA: 1.3/4.0 (1.0 best)
Jul 2014 – May 2016
M.Tech., Information & Communication Technology
DAIICT, Gandhinagar
- Cumulative GPA: 9.23/10.0
2014
B.E., Electronics & Communication Engineering
LD College of Engineering, Ahmedabad
Experience
Aug 2026 – Present
Senior Vice President – Machine Learning
Infocusp Innovations, Pune
- Promoted to lead the machine learning practice, with accountability for technical direction and delivery quality across the computer vision and LLM portfolio.
- Developing technical leads and strengthening evaluation and delivery standards across teams.
Jan 2025 – Jul 2026
Vice President – Machine Learning
Infocusp Innovations, Pune
- Led the computer vision group (10+ engineers) across multiple concurrent LLM and computer vision projects, converting open-ended client requirements into crisp project definitions, task breakdowns, and agreed success criteria.
- Led a 4-engineer team optimising MASt3R-SLAM for drone trajectory tracking on Jetson Orin NX, delivering a ~9× inference speedup (0.26 → 2.4 FPS) under tight compute and power budgets using TensorRT and ONNX, plus adaptive retrieval that skips redundant relocalisations during turns.
- Led a 5-engineer team building agentic AI systems for user researchers — agents that execute analysis workflows and generate reports — with LLM-as-a-judge evaluation harnesses as a first-class deliverable.
- Delivered a production hybrid multimodal retrieval system fusing vector-indexed image and text search with keyword search, for heterogeneous visual collections where off-the-shelf embeddings fail.
- Designed agentic health-protocol generation for a regulated domain with an auditable LLM review layer.
Aug 2023 – Jan 2025
Technical Lead – Machine Learning
Infocusp Innovations, Pune
- Fine-tuned TF Model Garden detection models for industrial defect inspection, achieving a 20% F1-score improvement on production data characterised by tiny targets, severe class imbalance, and few labelled failure examples; carried the work from data pipeline to customer POC.
- Built and led a 4-person R&D team on an LLM/RAG survey-analysis platform that summarises, tags, and reports on user-research data, cutting researcher time per project by 40%, validated against human-coded baselines.
- Translated peer-reviewed small-data techniques — transfer learning, attention-guided augmentation, one-shot detection — into repeatable delivery patterns for engagements where off-the-shelf models plateau.
Dec 2018 – Jun 2023
PhD Researcher
Friedrich-Alexander-Universität Erlangen-Nürnberg
- Pioneered ICC and ICC++, explainable feature learning methods leveraging human pose to uncover semantics and link iconography across image datasets.
- Enhanced pose estimation in ancient Greek vase paintings through style transfer and a perceptual metric, and improved one-shot object detection for heterogeneous artwork images using data contextualisation strategies.
- Improved classification performance for breast calcification analysis using histogram equalisation.
Jul 2016 – Nov 2018
Machine Learning Engineer
Infocusp Innovations, Ahmedabad
- Designed and deployed an enterprise candidate recommendation system processing 1M candidates daily to optimise hiring across multiple job positions.
- Led end-to-end development of a scalable hiring infrastructure using Python, PySpark and AWS, supporting concurrent multi-user access.
- Designed and implemented algorithms for reel, jerk, jigging and catch detection from fishing-rod sensor data — Python for development, C for deployment — with verification and validation testing.
- Contributed to tf-cnnvis, an open-source CNN visualisation tool.
Skills
Programming languages
AI/ML frameworks
- TensorFlow
- PyTorch
- Scikit-learn
- OpenCV
- Hugging Face
- Transformers
- ONNX
- TensorRT
Technologies
- Prompt engineering
- Docker
- CI/CD (GitHub Actions, Jenkins)
Tools
- Git
- VSCode
- Pandas
- NumPy
- IceVision
- LangChain
- LlamaIndex
- Tensorboard
- JupyterLab
- Streamlit
Databases
- ElasticSearch
- FAISS
- TFRecords
- Protobuf
Cloud
- AWS (EC2, S3, Lambda, SageMaker)
- GCP
Publications
21 peer-reviewed publications. The five most recent are below; the full list is on the research page.
AB Panambur, S Bhat, H Yu, P Madhu, S Bayer, A Maier. (2025). "Attention-guided erasing for enhanced transfer learning in breast abnormality classification." International Journal of Computer Assisted Radiology and Surgery.
M Zinnen, P Madhu, I Leemans, P Bell, A Hussian, H Tran, A Hürriyetoğlu, .... (2024). "Smelly, dense, and spreaded: The Object Detection for Olfactory References (ODOR) dataset." Expert Systems with Applications 255, 124576.
AB Panambur, P Madhu, S Bayer, A Maier. (2024). "Enhancing downstream classification of breast abnormalities in contrast enhanced spectral mammography using a neighborhood representation loss." Medical Imaging 2024: Computer-Aided Diagnosis 12927, 77-84.
AB Panambur, H Yu, S Bhat, P Madhu, S Bayer, A Maier. (2024). "Attention-guided Erasing: Novel Augmentation Method for Enhancing Downstream Breast Density Classification." International Journal of Computer Assisted Radiology and Surgery.
M Zinnen, P Madhu, P Bell, A Maier, V Christlein. (2023). "Transfer Learning for Olfactory Object Detection." arXiv preprint arXiv:2301.09906.
Talks
June 18, 2018
Tutorial at DAIICT, Gandhinagar, Gujarat, India
February 19, 2020
Talk at FAU Erlangen Nuernberg, Department of Pattern Recognition, Erlangen, Germany
February 19, 2020
Talk at FAU Erlangen Nuernberg, Department of Pattern Recognition, Erlangen, Germany
Teaching
