Vol. 01 / Full Stack AI
Full Stack AI Engineer • Currently @ Qure.ai
$300K+ ARRenterprise platform I architected
26s → <5sclinical AI pipeline latency at Qure.ai
Chapter
Latest from the Field Log
I called PyTorch's .backward() a thousand times before I understood what it was actually doing, so I built a tiny version to find out.
Qure.ai — Research & Optimization
Bangalore, India (Remote)
Foundation Industries (YC X25) — Full Stack & DevOps
San Francisco, USA (Remote)
Metzev
Georgia, USA (Remote)
unstudio.ai — Research & Backend
Gurgaon, India (Remote)
Styldod Inc. — Research & Optimization
Bangalore, India (Remote)
LIPG, IIT Indore
Indore, India
Led team of 5 to optimize LoRA-based try-on system for FLUX, achieving 2.5× faster boot-up and 98.7 PSNR on A40/H100 GPUs
Led team of 5–7 to fine-tune LLaMA, Gemma, Qwen & Phi models for complex SQL query tasks; achieved 90% accuracy
Latest
Multimodal AI system for soybean disease classification using vision-language models and attention-based reasoning, achieving 95% accuracy on real-world data. Integrated RAG pipeline for symptom-based remedy suggestions from damaged leaf images.
13th Inter IIT-Tech Meet, IIT Bombay
Production-grade multi-pipeline RAG with three specialized architectures (HybridFL, FLARE, DRAGIN) for legal and financial domains, with real-time knowledge graph updates via Pathway. Achieved 15–20% performance gains over GPT-4 and open-source LLMs on custom benchmarks; Docker-microservices, separate vector stores, and Streamlit production UI.
12th Inter IIT-Tech Meet, IIT Madras • Bronze (Rank 3 / 22+ IITs)
Dual-task ML pipeline: tweet engagement prediction at 96% accuracy and content generation with fine-tuned LLaMA. Scalable multimodal data processing for 1M+ heterogeneous records (text, image, audio, video).
Automated video editing pipeline achieving 50-60s processing for 5s UHD videos. Complete pipeline with AnimateDiff and ControlNet, evaluated using SOTA benchmarks.
Complete Django-based mess and hostel allocation system for IIT Indore. Engineered full-stack solution with comprehensive documentation and security measures.
Developed notebooks to standardize initial model setup and workflow testing.
Debugged and fixed errors in the LoRA training script, enabling fine-tuning of models.
Sustainable Cities and Society, Vol. 130, 2025
B. Kaurav, S. S. Dar, A. Jain, C. S. Raghaw, M. Z. U. Rehman, N. Kumar
Developed an explainable deep neural network architecture incorporating frequency-aware channel attention and spatial refinement mechanisms for accurate flood prediction in urban environments, contributing to sustainable city planning and disaster management.
Indian Institute of Technology Indore
GPA: 8.66 / 10 (current)
Inter IIT Tech Meet 12.0 — Bronze medal, Behaviour Simulation Challenge (2023)
All India Rank 1083 (out of 150k+ applicants, 2022)
All India Rank 5995 (out of 1M+ applicants, 2022)
Among top 300 candidates for the India team