AI Research and Development · Bengaluru

Shreyanth H G
AI Researcher and Developer

I design and publish parameter-efficient deep learning methods — and deploy them in production. My work spans low-resource LLM fine-tuning, computer vision, and real-time edge AI systems, with peer-reviewed research in MDPI's Computers journal.

LLM Fine-TuningEdge AIComputer VisionGenerative AI
scroll
Published in Computers (MDPI) // 56.3% faster fine-tuning // 99% of full-model accuracy with 40% of parameters // LLMs · ViTs · GANs // 1.2s edge inference on ESP32 // 5+ engineers mentored // 380+ students trained //
Profile

About

Overview

I am an AI/ML Engineer and researcher working at the intersection of academic research and production systems. At Accurate Info Solution in Bengaluru, I prototype research-driven machine learning solutions for client problems, with a focus on low-resource LLM fine-tuning and edge deployment.

My specialty is efficiency under constraint: adapter-based optimization, knowledge distillation, quantization, and deploying real-time computer vision on embedded hardware. This work culminated in SCALE, a peer-reviewed fine-tuning method published in MDPI's Computers.

Beyond engineering, I mentor junior engineers and regularly deliver workshops on Generative AI and data science to university audiences.

At a Glance

Role
AI/ML Engineer, Accurate Info Solution
Location
Bengaluru, India
Education
M.Sc. AI & ML (LJMU) + 2 PG programs + B.E. CS
Publication
Computers (MDPI), 2026 — SCALE
Focus areas
PEFT, Edge AI, CV, Generative AI
Core stack
PyTorch, Hugging Face, ONNX, OpenCV
Languages
English, Kannada, Hindi
0%
training-time reduction (SCALE vs. full fine-tuning)
0%
GPU memory saved in production fine-tuning workloads
0+
students trained through workshops and keynotes
0s
edge inference latency achieved on ESP32-CAM
Experience

Professional Experience

Aug 2023 — Present

AI/ML Engineer Current

Accurate Info Solution · Bengaluru
  • Spearheaded AI solution prototyping in a startup environment, delivering research-driven ML models to address client challenges.
  • Reduced LLM fine-tuning time by 40% and GPU memory usage by 35% through SCALE optimization on GPT-2 and BERT.
  • Designed modular ML pipelines and inference simulations for edge deployment using ESP32 and camera-based CV systems.
  • Built CV-based health monitoring tools, including facial stress recognition, and classification models for signal detection.
  • Mentored 5+ interns, resulting in 3 working GenAI/AIoT proofs-of-concept and consistent project delivery.
Feb 2022 — May 2023

Jr. Data Scientist (Trainee)

TuringMinds.AI · Remote
  • Developed a neural network from scratch in pure Python with manual hyperparameter tuning, achieving 88% accuracy on synthetic classification tasks.
  • Contributed to a No-Code AI platform, enhancing ML accessibility through modular plug-in logic.
  • Built dashboards and data visualizations for Mercedes-Benz and NASA datasets, enabling actionable analytics.
  • Designed data preprocessing workflows to handle sparsity, outliers, and transformation pipelines.
Capabilities

Technical Skills

ML / DL Frameworks

PyTorchTensorFlowKerasScikit-learnHugging FaceONNXTimm

LLM Fine-Tuning & Optimization

LoRAQLoRASCALEGPT-2BERTT5Mixed PrecisionMemory-Efficient TuningLayer Activation Scoring

Computer Vision

YOLOv12OpenCVViTCNNsGANsGrad-CAMCustom Detection Models

Edge AI & AIoT

ESP32 / ESP8266Arduino UnoSensor IntegrationServo FeedbackTorchScriptEdge InferencePyFirmata

NLP & Generative AI

RAGSentiment AnalysisQA ModelsTransformer Fine-TuningLSTM / GRUPrompt Engineering

Data & Tooling

Pandas / NumPyWandBTensorBoardGit / GitHubStreamlitDjangoSQLite
Selected Work

Projects

LLM OPTIMIZATION · PUBLISHED

SCALE — Adaptive Adapter Fine-Tuning Engine

Profiling-based adapter injection that selectively fine-tunes only the most critical transformer layers. Supports encoder-only, decoder-only, and seq2seq architectures. Published in Computers (MDPI), 2026.

−56.3%training time
99%accuracy retained
40%of parameters
GENERATIVE AI

Smart Document Retrieval System using RAG

A generative AI document assistant built on Retrieval-Augmented Generation, enabling sub-second retrieval and question answering over document collections.

<1sretrieval + QA
RAGarchitecture
COMPUTER VISION

Content-Based Image Retrieval with Feature Fusion

A hybrid CBIR system integrating deep InceptionV3 features with hand-crafted texture descriptors for super-resolution images, achieving high precision and F-measure on the VISTEX and STEX datasets.

Highprecision / F-measure
VISTEX+ STEX benchmarks
EDGE AI

Edge-AIoT Real-Time Pipeline

A real-time, sensor-triggered AI pipeline on ESP32-CAM with GPIO-controlled servos, demonstrating practical on-device inference under tight hardware constraints.

1.2sinference latency
ESP32-CAM deployment
Education

Academic Background

2024 – 2025

M.Sc. in Artificial Intelligence & Machine Learning

Liverpool John Moores University

2023 – 2024

Postgraduate Program in AI & Machine Learning

IIIT Bangalore

2021 – 2023

Postgraduate Program in Computational Data Science

Case Western Reserve University

2018 – 2022

B.E. in Computer Science

East West Institute of Technology, Bangalore

Leadership

Talks & Mentorship

290 students

GenAI Workshop — BGSCET

Conducted a full workshop covering transformers, prompt engineering, and real-world Generative AI applications.

90 students

Keynote — RRCE Data Science Workshop

Keynote presentation on statistical modeling and applied exploratory data analysis.

5+ interns

Mentorship — Accurate Info Solution

Mentored interns to deliver 3 proof-of-concept models in edge computer vision and low-resource LLM tuning.

Jury member

National-Level Hackathon

Served as a jury member for a national-level hackathon, evaluating AI and software solutions from participating teams.

Peer-Reviewed Research

Published Research

My research focuses on parameter-efficient fine-tuning of transformer models — making state-of-the-art NLP accessible in resource-constrained environments.

Computers · MDPI Open Access · Scopus Indexed Published March 2026

A Resource-Efficient Approach to Fine-Tuning a BERT-Base Model for Sentiment Analysis

Abdullah M. Basahel, Shreyanth H. Giriyappa, Furqan Alam, Tahani Saleh Mohammed Alnazzawi, Saqib Qamar, Adnan Ahmed Abi Sen

Computers 2026, 15(3), 159 · DOI: 10.3390/computers15030159 · Section: AI-Driven Innovations

This paper introduces SCALE (Selective Critical Adapter Layer Efficiency), a novel parameter-efficient fine-tuning method. Rather than adapting every transformer layer uniformly, SCALE dynamically profiles layers using activation magnitude and attention entropy, then injects lightweight adapter modules only into the most influential layers. Benchmarked against LoRA, AdaLoRA, and BitFit across five sentiment analysis datasets (IMDB, SST-2, Yelp, TweetEval, XLive), SCALE consistently delivers superior accuracy, F1-score, and AUC while substantially reducing training cost.

0% reduction in training time vs. full fine-tuning
0% of full-model accuracy retained
0% of parameters used (61.4% fewer trainable)
+0% accuracy over LoRA on IMDB
01

Dynamic Layer Profiling

A novel profiling technique that scores each transformer layer via activation magnitude and attention entropy, identifying the layers most relevant to the downstream task before any training begins.

02

Selective Adapter Injection

Lightweight residual adapter modules are inserted only into the top-k highest-scoring layers, preserving pretrained representations while enabling targeted, interpretable adaptation.

03

Validated Efficiency

Rigorous 5-fold evaluation across binary and multi-class benchmarks, with a full ablation study against uniform and adaptive layer-selection strategies under identical training conditions.

Research interests: Parameter-Efficient Fine-Tuning Knowledge Distillation Edge-Optimized Deep Learning Medical Imaging AI Multimodal Learning Model Compression
Contact

Let's work together.

Open to research collaborations, applied AI roles, and challenging engineering problems. Reach out through any of the channels below.