JEEVAKAMAL K RAI Engineer.
I build and deploy production-grade intelligent systems. Specializing in scalable architectures, hybrid RAG pipelines, and edge optimization.
Industry Experience
AI Trainee(Product Development)
UVS Cube Infotech
Mar 2026 – May 2026 | Madurai
Designed and deployed production-grade healthcare AI systems with deterministic safety guarantees and low-latency hybrid retrieval pipelines.
Medical Triage Chatbot
The Architecture
Built a hybrid RAG pipeline combining FAISS (dense) and BM25 (sparse) retrieval over PubMedBERT embeddings. Integrated with FHIR R4/ABDM for real-time patient history injection into LLM prompts.
Engineering Metrics
- Engineered a deterministic severity scoring engine (0–100) with red-flag detection.
- Implemented Triple-Gate output safety guaranteeing zero false diagnostic statements.
Voice Assistant Pipeline
The Architecture
Developed an end-to-end voice pipeline using Whisper for ASR, Llama-3.1 for NLU, and pyttsx3 for TTS. Enabled voice-to-UI control for hands-free healthcare app navigation.
Deployment & Infrastructure
- Deployed both systems to production via Docker + NGINX.
- Automated deployment pipelines using GitHub Actions CI/CD.
- Implemented Redis session caching and Groq→Gemini LLM failover strategies.
Advanced Architectures
Production-grade model architectures and cyber-physical systems.
Cyber-Physical System
Digital Twin for Urban Microgrid Resilience
Designed a cyber-physical digital twin integrating CNN-BiLSTM load forecasting, Model Predictive Control (MPC) for dispatch optimization, and Extended Kalman Filter (EKF) for real-time state estimation.
Load Forecasting
CNN-BiLSTM with 96% R²
Control System
MPC + EKF real-time loop
Communication
MQTT pub/sub pipeline
Decentralized Architecture
FedRetinaNet: Defect Detection
Developed a privacy-preserving federated RetinaNet framework for insulator defect detection. Achieved mAP@50 of 0.91, retaining 97% of centralized performance while preserving edge data privacy.
Physics-Constrained ML
EV Range Prediction using PINNs
Engineered a Physics-Informed Neural Network (PINN) layered with LSTM and Attention mechanisms to predict EV range under non-stationary real-world traffic with SUMO simulation.
Affective Visual Intelligence
AVIS: Emotion Recognition & Image Captioning
Built a dual-stream Affective Visual Intelligence System combining EfficientNet + MediaPipe CNN for 7-class emotion recognition (63% accuracy) with BLIP and LangGraph AI for empathetic captioning.
Applied AI Projects
Agentic Knowledge Graph
ResearchOS: Enterprise GraphRAG Platform
Built an autonomous AI research operating system using Multimodal RAG and LangGraph agents. Engineered a tri-service microservice architecture (Nginx, FastAPI, Neo4j) to extract entities from ArXiv papers and construct explicit, traversable Knowledge Graphs for hallucination-free reasoning.
Model Engineering & Optimization
Efficient English-Tamil Translation
Fine-tuned a LLaMA-3.1-8B model specifically for English-to-Tamil translation. Utilized 4-bit LoRA (Low-Rank Adaptation) and Unsloth to drastically reduce memory footprint and training time while maintaining high BLEU scores.
Research Works
HPC & Edge Deployment
Heterogeneous Satellite Imaging Benchmark
Developed a comprehensive benchmark for heterogeneous satellite imaging systems, integrating High Performance Computing (HPC) with Computer Vision (CV) to optimize edge deployment and processing performance.
Computer Vision & IoT
Shelf-Life Prediction of Fruits
Designed an automated shelf-life prediction system for fruits using advanced Deep Learning models deployed alongside IoT sensor networks for real-time environmental monitoring and visual analysis.
Technical Expertise
AI Systems Engineer — Bridging AI & Systems
Specializing in building end-to-end AI pipelines, from training custom architectures and fine-tuning large language models to deploying robust, high-performance backends for mission-critical and domain-specific applications.
