Phone: +91-9910050802
Email: saggutaranpreetsingh@gmail.com
GitHub: github.com/TaranpreetSinghSaggu
I am an AI engineer focused on turning machine learning and LLM ideas into reliable products, from multimodal experimentation to agentic RAG systems and full-stack delivery.
Snapshot
Quick overview of my current focus and strengths.
About
My work sits at the intersection of deep learning research, intelligent document systems, and product engineering. I enjoy building from first principles, benchmarking what matters, and carrying ideas through APIs, deployment, and user-facing delivery.
Research
Multimodal models, deepfake detection, and custom neural network implementations.
Systems
RAG pipelines, vector retrieval, Redis-backed workflows, and production-minded evaluation.
Delivery
Full-stack product ownership spanning frontend, backend APIs, authentication, and cloud deployment.
Featured Projects
These projects highlight my mix of LLM systems thinking, research rigor, and product ownership.
Local LLM Systems
Built an Ollama-based local code generation pipeline with custom ModelFile configuration on local .gguf models.
Developed a Jupyter magic command (%%prompt) to stream responses and auto-insert generated Python code into notebook cells.
Implemented an end-to-end workflow with CLI testing and notebook integration for interactive code generation and execution.
LLM Fine-Tuning
Built a comprehensive experimentation repository for LLM fine-tuning across multiple datasets.
Explored both single-dataset and multi-dataset fine-tuning strategies by combining datasets to improve generalization and downstream performance.
Implemented and evaluated dataset curation, parameter tuning, and training optimization techniques for improved outcomes.
Agentic RAG
Built a financial document intelligence platform to analyze Ministry of Finance and RBI policy documents.
Indexed 2024 monthly financial reports with vector embeddings to enable semantic search over government publications.
Implemented agentic workflows with Redis caching and vector search to improve query efficiency and response latency.
Research Systems
Explored Liquid Neural Networks (LNNs), a continuous-time recurrent architecture using Liquid Time Constants for temporal adaptability.
Reproduced the Liquid Time-Constant (LTC) network and implemented hand gesture segmentation from scratch using 32-dimensional time-series motion data.
Built custom LTC cells in PyTorch and benchmarked against LSTM, achieving 54.6% validation accuracy vs 45.0% with a 23% lower generalization gap.
Experience
My experience reflects a mix of experimental depth and shipping responsibility.
Gurmat Darbar (Community Web Application)
Sep 2025 - Present
IIT Mandi
Aug 2025 - Dec 2025
Education
A dual-degree path that supports both theory-heavy learning and practical engineering work.
BS in Data Science and Applications
2022 - 2027 | CGPA: 7.0
B.Tech in Computer Science (AI & ML)
2023 - 2027 | CGPA: 9.0
Skills
Organized around how I design, build, and ship practical AI systems.
Resume + Contact
If you want to collaborate, discuss ideas, or explore my projects in more detail, feel free to connect.