Gruhesh Sri Sai Karthik Kurra
I build AI systems from scratch and ship them to the real world 🧠🤖🚀
About
Independent AI/ML researcher and builder, specialising in building AI systems end-to-end: from implementing transformer architectures from scratch to deploying on-device deepfake detectors. Expertise spans LLM optimisation, computer vision, neural network compression, and synthetic data generation.
As a Research Intern at IIIT Hyderabad, I developed a Layout Preserving Document Translation system achieving 92% accuracy in layout detection with multi-language OCR support. I have published 7 research papers in international conferences and journals (IEEE, IJNRD, Zenodo) covering neural network compression, deepfake detection, and novel word representation systems.
Certified AWS Cloud Practitioner, AWS Solutions Architect, TensorFlow Developer, and WSO2 Micro Integrator Developer. Experienced across the full AI/ML stack — from building LLMs from scratch to deploying production systems on cloud infrastructure. I write an ML-from-scratch blog at blogs.gruheshkurra.com.
Education
B.Tech in Computer Science and Engineering
KL University, Hyderabad, India · August 2021 – May 2025 · CGPA: 9.72 / 10
IELTS Academic · Overall Band Score: 7.5 · C1 Advanced
Research Publications
-
RADAR: Reasoning-Augmented Deepfake Artifact Recognition via Multi-Branch Evidence Aggregation Under Review IJECE · April 2026Novel architecture modelling deepfake detection as structured evidence aggregation through three complementary branches — STAM (skin texture), BADM (boundary aliasing), AADM (GAN fingerprints) — with a Reasoning Module using recurrent cross-attention for cross-dataset generalisation.
-
Morphology-Aware Nested Character Embeddings for Word Representation Accepted IEEE ICCCMLA 2025 · 2025Novel character-level representation modelling words as nested character sequences via bidirectional LSTMs and Transformers. Achieved +2–8% improvement on English classification tasks and +58.3% improvement on German NER over baselines.
-
BIE: Bit-Index Encoding for Efficient Neural Network Weight Compression Zenodo · September 2025Novel compression framework representing neural network weights through bit-level indexing, achieving compression ratios up to 40× with MSE < 10⁻⁶ and minimal accuracy degradation. Includes complete open-source implementation with optimised sparse kernels.
-
Hybrid RAG-Enhanced Deepfake Detection: Combining Retrieval-Augmented Generation with Visual Inconsistency Analysis Zenodo Preprint · August 2024Privacy-first deepfake detection combining RAG-based semantic retrieval with visual anomaly detection and uncertainty-aware classification for enhanced accuracy and interpretability.
-
Voice-Activated AI for Seamless Computer Interaction IJNRD · January 2025End-to-end voice command system with intent classification and system control for seamless human–computer interaction.
-
Global Remote RAM Sharing: A Novel Framework for Distributed Computational Systems IJNRD · December 2024Innovative framework optimising memory utilisation in distributed systems through global remote RAM sharing mechanisms for compute-constrained environments.
-
Dynamic Auto-Finetuning of Language Models Based on Confidence-Driven Knowledge Integration IJNRD · November 2024Confidence score triggers selective fine-tuning without full retraining — novel approach for dynamically adapting language models using confidence metrics and knowledge integration strategies.
Experience
2025
- Led development of an AI-driven integration framework connecting enterprise applications with cybersecurity services via standardised APIs and intelligent orchestration.
- Architected agent-based system using Azure OpenAI for automated schema mapping and adaptive connection strategies with performance learning from historical patterns.
- Built full-stack application using React 18, Vite, TailwindCSS + FastAPI backend with WebSocket integration for real-time monitoring and agent coordination.
- Integrated Azure Active Directory via Microsoft Graph API for application discovery and automated metadata management.
Jun 2025
- Led development of a Layout Preserving Document Translation system: Vision Transformer-based layout detection (92% accuracy), character metrics-based font detection, K-means colour palette extraction (88% accuracy).
- Built FastAPI backend integrating Tesseract OCR for multi-language text extraction (English, Hindi, Telugu, German) while preserving document formatting.
- Presented at IIIT Hyderabad research expo (May 2025) with live demonstrations to faculty and industry visitors.
Jun 2025
- Developed and deployed RESTful APIs for PLDT (Philippines' largest telecoms provider) using WSO2 Integration Studio for ticketing system integration.
- Managed production ticket systems with secure credential management through GitLab and Choreo cloud platform with vault-based secrets management.
- Engaged directly with international clients under strict SLAs to resolve production issues in environments with sparse framework documentation.
2024
- Completed intensive 6-month apprenticeship covering neural architecture design, LLM fundamentals, prompt engineering, model fine-tuning, quantisation, and deployment strategies.
- Worked on research projects in document analysis, OCR systems, and computer vision under faculty mentorship.
Jan 2025
- Founded a synthetic data platform producing tabular, image, and text data using GANs and VAEs to address dataset scarcity in ML training.
- Designed meta-controller architecture where a mother model orchestrates 200–1000 specialised child models to generate millions of rows without duplication.
- Implemented prototype with transformer-based PII detection (98.5% accuracy) and 95% statistical similarity between synthetic and real data.
2023
- Exploratory data analysis, statistical modelling, and ML model development using Python (pandas, NumPy, scikit-learn, Matplotlib).
- Feature engineering and model evaluation on real-world datasets in retail analytics and customer segmentation.
Selected Projects
EfficientNet-B1 trained on 25k images; 98.62% accuracy in 55 min on M4 MacBook. Converted to 26MB Core ML model for real-time on-device iOS inference in <500ms. Full-stack FastAPI + React web app.
"Attention Is All You Need" replicated end-to-end — full encoder-decoder with BLEU benchmarks. Separately implemented GPT-2 (124M params) from scratch with BPE tokeniser and full training loop.
Novel neural weight compression achieving 40× ratio with MSE < 10⁻⁶. Published on Zenodo (2025) with complete open-source implementation including optimised sparse kernels and benchmarking suite.
PubMedBERT clinical NLP pipeline for trial matching. Achieved 99.5% accuracy on biomedical similarity tasks using domain-specific pre-trained embeddings and FAISS vector retrieval.
Autonomous agent that reads GitHub issues, writes fix patches, and opens PRs — zero server needed. Runs entirely via GitHub Actions with Python orchestration and LLM-driven code editing.
GAN trained on MNIST with multiple architectures (vanilla GAN, DCGAN). Full training pipeline with loss curves, generation quality metrics, and latent space interpolation.
DDPM implemented from math to generated images. ~130K parameter model trained end-to-end with full noise schedule, denoising UNet, and sample generation pipeline.
Privacy-first deepfake detection combining retrieval-augmented generation with visual inconsistency analysis. Uncertainty-aware classifier for enhanced interpretability. Published on Zenodo 2024.
Implementations of Linear Regression, Logistic Regression, KNN, Naive Bayes, Decision Trees, Random Forest, SVM, K-Means, t-SNE, UMAP — all from math up with no sklearn wrappers.
Vision Transformer layout detection (92% accuracy) + multi-language OCR (English, Hindi, Telugu, German) preserving fonts, colours, and document structure. Research project at IIITH.
Technical Skills
Programming Languages
AI / ML Frameworks
AI / ML Specialisations
Computer Vision
iOS / Mobile
Web & Backend
Cloud & DevOps
Databases & Data Science
Certifications
Awards & Honours
Developed a comprehensive diet management platform in a 4-hour hackathon featuring personalised meal planning, calorie tracking, and nutritional recommendations. Competed against 50+ teams.
Created an Arduino-based smart switchboard for IoT home automation with real-time energy monitoring and mobile app control. Recognised for innovative approach to energy efficiency and user-centric design.
Awarded for technical leadership as Technical Lead. Mentored 20+ juniors in web development and ML, organised workshops for 100+ students, and coordinated 15+ technical events.
Contact
Email: gruheshkurra2@gmail.com
Phone: +91 8008258065
Location: Hyderabad, India
Twitter/X · GitHub · LinkedIn · Hugging Face · Blog · Email
Open to collaborating on: Transformer Research · LLM Development · iOS On-Device AI · Vision-Language Models · Neural Compression · Agentic Systems