Saurabh Kumar Singh
AI/ML Engineer — Computer Vision · Vision-Language Models · Generative AI
+91-8210746826 · saurabhsingh802213@gmail.com · Sasaram, Bihar — open to relocation / remote
linkedin.com/in/saurabh-kumar-singh-b02b93278/ · github.com/saurabh23011
Summary
AI/ML Engineer (M.Sc. AI & ML, IIIT Lucknow) with 2+ years of hands-on work in computer vision, generative AI and multimodal model evaluation; currently evaluating Vision-Language Models at Snorkel AI. Built a monocular-depth 2D-to-3D video pipeline, a generative textile-design system and an OCR invoice-extraction pipeline, backed by a B.Sc. in Mathematics.
Technical Skills
- Languages
- Python, SQL, C, C++, JavaScript
- ML / Deep Learning
- PyTorch, TensorFlow, Keras, Hugging Face Transformers, Scikit-learn, CNNs, Transformers, Transfer Learning, Hyperparameter Tuning, Model Evaluation & Error Analysis
- Computer Vision
- OpenCV, Object Detection, Image Segmentation, Monocular Depth Estimation (MiDaS/DPT), Depth-Image-Based Rendering, OCR (Tesseract), Image Preprocessing & Enhancement
- Generative AI / LLMs
- Diffusion & Image-Generation Workflows, Vision-Language Models, Prompt Engineering, RAG, LLM Evaluation, Hallucination Detection, Instruction & Preference Data Curation
- Data & Deployment
- NumPy, Pandas, Matplotlib, Seaborn, FastAPI, Flask, Streamlit, REST APIs, Docker, Git, Linux, MySQL, MongoDB, CUDA
Experience
AI Specialist — Computer Vision & Multimodal Evaluation (Contract)
Aug 2026 — PresentSnorkel AI
Remote
- Evaluate Vision-Language Model outputs on image and video tasks, catching hallucinations, spatial-reasoning errors and wrong object/relationship claims, and write corrective annotations to improve the models.
- Analyse detection, segmentation, depth, geometry and image-comparison tasks, breaking complex scenes into parts to pinpoint where a model fails, and build multimodal training/evaluation data for visual reasoning.
Freelance AI/ML & AI Evaluation Specialist
2024 — PresentContract work for AI training and evaluation platforms
Remote
- Author expert-level evaluation tasks, rubrics and reference solutions for LLMs and AI agents, including image-based reasoning, mathematics and agentic tool-use scenarios.
- Grade model responses against rubrics, flag reasoning and factual errors, and write preference/correction data for training.
Lecturer — Computer Science
Jul 2025 — Dec 2025RSR Rungta College of Engineering and Technology (RSR-RCET), Bhilai
On-site
- Taught Image Processing, Cybersecurity, Cryptography and C to UG/PG students; mentored AI capstone projects.
- Built an academic timetable system that automatically resolves faculty, classroom and period clashes for about 35 faculty, replacing manual spreadsheets.
Junior Computer Vision Engineer
Feb 2025 — Jun 2025CLAW Legaltech Private Limited
Remote
- Built a generative-AI textile design system that creates custom fashion prints and patterns from user inputs, using PyTorch and image-generation models.
- Added preprocessing, enhancement and style-variation controls, turning the research prototype into a product feature, and iterated with product stakeholders on quality and consistency.
Junior Data Scientist
Jan 2024 — Jun 2024Logiciel Analytics Private Limited
Remote
- Built an end-to-end OCR pipeline (Python, Tesseract, OpenCV, Pandas) that turns scanned invoices into structured, queryable records.
- Improved recognition accuracy with noise removal, adaptive thresholding, deskewing and contrast enhancement; automated a manual data-entry workflow.
Projects
2D-to-3D Video Conversion using Monocular Depth Estimation
M.Sc. DissertationPython, PyTorch, OpenCV, NumPy, CUDA
- Built a pipeline that converts standard 2D video into stereoscopic 3D using monocular depth estimation and depth-image-based rendering (DIBR).
- Benchmarked MiDaS DPT-Large vs DPT-Hybrid (depth quality vs speed); synthesised left/right views with pixel warping and hole-filling, and used GPU acceleration and temporal smoothing to reduce flicker.
Generative Textile Design System
CLAW LegaltechPyTorch, Diffusion Models, OpenCV, Python
- Generative-AI system that creates custom fashion prints and patterns from user inputs.
- Preprocessing, enhancement and style-variation controls took the prototype to a shipped product feature.
OCR Invoice Extraction Pipeline
Logiciel AnalyticsTesseract, OpenCV, Pandas, Python
- End-to-end pipeline turning scanned invoices into structured, queryable records.
- Noise removal, adaptive thresholding, deskewing and contrast enhancement lifted recognition accuracy and removed a manual data-entry step.
Education
Certifications & Achievements
Qualified CMI & KSOM PG Mathematics entrance | Microsoft Learn AI Skills Challenge | HackerRank Software Engineer Intern | HackerRank SQL (Advanced)