AI Engineer & Researcher. Motivated by the hardest, most challenging and multidisciplinary problems — from safe autonomous systems and 3D perception to LLMs, data science and quantitative research. MSc Artificial Intelligence @ University of Manchester.
I'm an AI engineer and researcher driven by a passion for challenging, multidisciplinary problems. During my B.Tech at VIT Chennai, I built one of the most extensive research records in my cohort — authoring 5 publications across IEEE Access, Elsevier and international conferences, filing 2 patents, conducting funded research in India and South Korea, and presenting at conferences as far afield as Minsk, Belarus.
My work spans an unusually wide range: safe reinforcement learning for autonomous vehicles, semantic segmentation of satellite imagery, 3D point cloud classification for AR/VR, robotic arm design, voice-enabled navigation, agri-robotics and deepfake detection. Most recently, I built a full RAG pipeline for East Asian cuisine Q&A as part of my MSc coursework.
I bring the same depth and rigour whether building LLM/RAG pipelines, quantitative models, autonomous systems, or data-driven products. Currently based in Manchester, actively seeking roles in AI/ML engineering, data analysis and quantitative research.
K. Pradeep, K.P. Vijayakumar, K. Palani Thanaraj, Rohan Inamdar, S. Kavin Sundarr
Patent · India #202441087748 · Published Nov 2024
A Robotic Hand for Automation of Tasks
Rohan Inamdar, S. Kavin Sundarr, Ajeyprasaath KB, Vetrivelan P, Krishna Kumba, Patri Upendar
04
Projects
NLP · RAG · LLM2025
RAG Culinary Assistant
End-to-end retrieval-augmented generation system for East Asian cuisine Q&A. Built from scratch across 6 Jupyter notebooks — data collection, FAISS ingestion, benchmarking, inference, evaluation and ablation studies. Gradio UI for demo day.
Lightweight CNN with non-local attention for 3D object classification. Outperforms SOTA by 1.4%. KEITI-funded internship at Jeonbuk National University. Published in IEEE Access.
Hybrid model fusing UNet decoder with Swin Transformer encoder for satellite image segmentation. Outperforms DeepLab-v3+, UNet and SegNet. Published at CVIP 2024.
Voice-commanded autonomous nav with 100+ commands, LiDAR + camera, 1.2s avg response across 5 map configs using ROS2 Nav2 stack. Paper published at AIITA 2024, Springer.
Real-time deepfake detection at 82% accuracy. Dedicated GRU model for video temporality combined with CNN image classifier. Trained across diverse multi-source datasets.
Sentence-level lip reading at 97% accuracy using 3D-CNN and BiLSTM. Deployed as a full Django web application for real-time subtitle generation from lip movements.
TD3-based RL agent in ROS2 Gazebo 11 for autonomous obstacle avoidance. 82% navigation accuracy. Full training pipeline monitored with PyTorch and TensorBoard.
World Bank & ICAR funded autonomous agri-robot for pest and disease detection. Voice-enabled robotic arm with navigation deployed in field conditions under NAHEP programme.
MSc AI researcher with hands-on experience in statistical modelling, data pipelines, NLP analytics and ML-driven insight generation. I build end-to-end systems that go from raw data to actionable output.
My analytical foundation was built across 5 published research projects requiring rigorous data collection, statistical evaluation, and result interpretation. From benchmarking RAG retrieval pipelines to evaluating segmentation model performance against multiple baselines, I've consistently worked in data-heavy environments.
My MSc in AI at Manchester has deepened my grasp of statistical learning, experimental design, and evaluation methodology. I'm comfortable working with messy, real-world datasets — I've done it in agriculture, robotics sensor data, NLP corpora, and satellite imagery.
I combine the rigour of a researcher with the practicality of an engineer — I don't just analyse data, I build the pipelines that make analysis repeatable and scalable.
02
Relevant Projects
NLP · Data Pipeline · Evaluation2025
RAG Culinary Assistant — Full Analytics Pipeline
Built a complete data pipeline from scraping to evaluation. Implemented benchmarking notebooks to compare retrieval strategies, measured precision/recall metrics, and conducted ablation studies to quantify the impact of each system component on output quality.
Comparative analysis of sentiment models. Evaluated multi-head attention BiLSTM against baseline approaches using standard NLP metrics across multiple benchmark datasets.
Rigorous comparative evaluation of segmentation models across aerial image datasets. Statistical analysis of performance gains over DeepLab-v3+, UNet, and SegNet baselines. Published at CVIP 2024.
Systematic literature review synthesising findings from 100+ papers on safe RL for autonomous vehicle control. Data-driven analysis of algorithm performance across environments. Published in Elsevier E-Prime.
MSc AI researcher with a strong foundation in mathematical modelling, optimisation, statistical learning, and algorithmic problem-solving — the core toolkit of quantitative analysis.
Quantitative work is fundamentally about building rigorous models under uncertainty and optimising decisions over time — which is exactly what I've done in my research. My work on safe reinforcement learning for autonomous vehicles required designing reward functions, handling stochastic state transitions, and imposing hard constraints on policy behaviour.
My MSc modules span Knowledge Representation, Probabilistic Reasoning, and NLP — all grounded in mathematical formalism. I have strong experience with optimisation in high-dimensional spaces, both in theory and implementation.
I'm drawn to quant roles because they demand the same mindset I've developed as a researcher: formulate the problem precisely, model it rigorously, validate empirically.
02
Relevant Projects
RL · Optimisation · Published2024
Safe RL for Autonomous Vehicles — Constrained Optimisation
Researched and implemented constrained policy optimisation for safe trajectory planning in stochastic environments. Designed reward shaping functions with hard safety constraints. Published systematic review of RL algorithm performance in Elsevier E-Prime. Directly analogous to risk-constrained portfolio optimisation.
Designed ablation studies and benchmarking methodology to isolate the contribution of each RAG component. Statistical evaluation framework measuring precision, recall, and generation quality across configurations.
Lightweight 3D Classifier — Efficiency Optimisation
Designed a computationally efficient point cloud network using non-local attention, minimising model complexity while maximising accuracy. Beat SOTA by 1.4% — a direct trade-off optimisation problem. Published in IEEE Access.
MSc AI researcher with 5 publications, 2 patents and 10+ projects spanning NLP, computer vision, safe RL, and robotics. I build models end-to-end — from architecture design to deployed systems.
I've been building and publishing AI systems since my second year of undergrad. What separates me from most candidates is that my projects aren't just code — they're published research. Every major project I've worked on has been evaluated rigorously, compared against baselines, and documented to publication standard.
My technical range is broad: transformer architectures for vision and NLP, reinforcement learning in continuous action spaces, 3D point cloud processing, RAG pipeline engineering, and robotic systems with real-world hardware deployment. I'm comfortable from research paper to GitHub to deployed demo.
I bring research-grade rigour to engineering problems and engineering practicality to research questions — the combination that makes an effective ML engineer.
02
Research Experience
Fall 2023 – Present
SENSE Lab, VIT Chennai
Researcher — Safe RL & Semantic Segmentation
Researched safe trajectory planning for self-driving vehicles using constrained Reinforcement Learning
Developed hybrid UNet + Swin Transformer model for precise satellite image segmentation
Building a unified lightweight model covering aerial, ground and underwater datasets
Contributed to 2 published journal papers; mentored by Prof. Nitish Katal
Chennai, India
Summer 2024
Jeonbuk National University · KEITI Funded
Research Intern — 3D Object Classification
Funded by Korea Environmental Industry and Technology Institute
Designed lightweight Point Cloud Network for 3D classification in AR/VR
Outperformed SOTA baseline by 1.4% on benchmark datasets
South Korea
Winter – Summer 2024
SENSE Lab, VIT Chennai
Researcher — Robot Hand & Voice Navigation
Voice-enabled rerouting system for robot navigation failure recovery on TurtleBot3 / Nav2
Designed precision robot hand for fine manipulation — filed as Indian patent #202441087748
Chennai, India
03
Projects
RAG · LLM · NLP · 2025Active
RAG Culinary Assistant
End-to-end RAG system built from scratch across 6 Jupyter notebooks — FAISS-based vector ingestion, LLM inference, benchmarking against baseline approaches, evaluation metrics (precision/recall/BLEU), and ablation studies. Gradio UI for demo.
97% accuracy sentence-level lip reading. 3D convolution + bidirectional LSTM deployed as a full Django web application for real-time subtitle generation.