AI Engineer & Software Engineer specializing in production machine learning systems, schema-aware LLM/RAG microservices, robust backend APIs, and distributed cloud-connected IoT infrastructure. Undergraduate in B.Tech CSE (AI & ML) at Vellore Institute of Technology.
Organized across eight core engineering domains. Every technology listed is backed by verified production implementations, open-source pull requests, published patents, or validated research systems.
Deep learning, statistical modeling, computer vision, acoustic representations, and temporal forecasting.
End-to-end LLM orchestration, semantic vector retrieval, NL-to-SQL translation, and structured output parsing.
High-performance programming languages, data structures, modular architecture, and object-oriented principles.
Production API services, asynchronous handlers, authentication, rate limiting, and real-time streaming.
Cloud computing services, edge IoT deployments, hardware telemetry, and automated event workflows.
Relational schemas, real-time message brokers, analytical feature engineering, and document parsing.
Modern, responsive client interfaces integrated with backend microservices and secure database endpoints.
Version control workflows, API testing, interpretability suites, and empirical statistical validation.
Delivering high-throughput document intelligence, multimodal computer vision architectures, schema-aware NL-to-SQL translation, and clinical decision microservices across enterprise production environments.
Ranked #80 globally out of 43,587 international developers in GirlScript Summer of Code 2026. Merged 179 pull requests across 7 production repositories spanning autonomous AI agents, streaming backend APIs, and full-stack software tooling.
Contributed across full-stack repositories with production Next.js, Python, FastAPI, and Node.js architectures. Engineered multi-turn agent state management, validated structured JSON outputs against runtime schemas, built real-time Server-Sent Events (SSE) inference pipelines, and fortified API endpoints with concurrency locks, rate limiting, and automated CI test validation.
Implemented directed agent execution workflows, structured-output validation using strict Pydantic/Zod schemas, and session memory persistence across multi-turn agent interactions in Next.js and Python.
Engineered resilient Server-Sent Events (SSE) endpoints in FastAPI and Node.js for ultra-low latency token streaming and intermediate agent reasoning thought traces to client web applications.
Hardened backend API routes with JWT authentication guards, token-bucket rate limiters, concurrent transaction locks, unified error middleware, and cache consistency mechanisms across server actions.
Added automated integration test suites, input sanitization middleware against prompt injection vulnerabilities, and strict schema contracts across 7 public open-source repositories.
Architected the official NEXUS conversational AI chatbot system for student query resolution, event registrations, and campus technical workshops.
Mentored junior engineering members in experimental design, dataset benchmarking, code review standards, and reproducible ML pipeline workflows.
An applied cyber-physical engineering initiative combining distributed edge sensing, encrypted cloud telemetry, deep learning forecasting, and automated valve actuation across the Vellore Institute of Technology campus.
Led a 14-member multidisciplinary engineering team across 8 workstreams to build and deploy an end-to-end intelligent IoT sensing network, cloud ingestion pipeline, and ML-driven precision water delivery system.
Custom 3D-printed weather-resistant enclosures with 7 sensing channels each.
Captured over 90 continuous days at 15-minute telemetry intervals.
Derived from hardware architecture, telemetry pipelines, and predictive algorithms.
MQTT to AWS IoT Core pipeline with local edge caching fallback.
Reframing academic research as applied computational engineering: developing production-grade machine learning pipelines, formal sequence models, dataset harmonization suites, and empirical evaluation frameworks across international research collaborations.
SUNY Binghamton
Vellore Institute of Technology
QNeuroVellore Institute of TechnologyEnd-to-end systems combining machine learning algorithms, asynchronous backend services, spatial graphs, distributed message streaming, and hardware-in-the-loop validation.
Real-time autonomous vehicle policy training typically requires heavy C++ or GPU runtimes while operating as uninspectable black boxes, failing to provide safety engineers with live spatial decision boundaries or verifiable neural weights.
Engineered DriveGraph, a zero-dependency autonomous vehicle simulation platform in modern Vanilla JavaScript and HTML5 Canvas 2D running at 60 FPS in-browser; parses OpenStreetMap (OSM) GIS road graphs via Overpass API, computes Minkowski road surface envelopes, casts 15 LiDAR-like 360° ray beams, visualizes 2D hyperplane decision boundaries, and evolves sparse neural controllers.
Severe 3-phase load imbalance across electrical distribution networks causes catastrophic transformer overloads, 5–8% energy loss spikes, and line outages due to uncoordinated manual phase switching during peak hours.
Built a 3-layer edge-cloud control system connecting ESP32 household nodes, Raspberry Pi transformer controllers, and HQ analytics servers; integrated RS-485/Modbus, LoRa wireless mesh, and MQTT/TLS with autonomous edge decision-making, zero-crossing detection, relay interlocks, and dual-stage cryptographic authorization.
Commercial cloud AI tools transmit sensitive legal inquiries to external third-party servers, produce ungrounded hallucinations, and lack verifiable statutory citations required for criminal law research.
Architected LexBNS, an enterprise-grade, 100% offline Retrieval-Augmented Generation (RAG) and ReAct reasoning engine; pairs local Ollama models (Llama 3.1 8B / Mistral 7B) with dense FAISS vector search over statutory PDFs (Bharatiya Nyaya Sanhita 2023 & Constitution of India); containerized via Docker Compose with a Streamlit interface featuring expandable source citation drawers.
Early detection of affective and neurological biomarkers from voice is compromised by severe acoustic entanglement: speaker identity, recording devices, and background noise overwhelm subtle diagnostic prosodic signals.
Developed HEDER-Net (B.Tech Final Year Project-1), an end-to-end speech emotion disentanglement neural framework built on frozen emotion2vec+ acoustic representations; engineered a multi-module neural pipeline incorporating temporal attention, Bi-LSTM sequence encoders, VAE-style latent space purification, prototype learning, and Gradient Reversal Layer (GRL) adversaries under an 8-component joint multi-task loss with dynamic weighting.
Real-time automated spacecraft descent in uncertain physical simulation environments is safety-critical and highly susceptible to sample correlation, destabilizing thruster chattering, and policy divergence.
Engineered an optimized Deep Q-Network (DQN) framework for autonomous spacecraft guidance within Gymnasium LunarLander-v3; implemented Experience Replay (10^5 FIFO buffer) and Soft Target Network Synchronization (tau=0.001); conducted a multi-parameter sensitivity grid across learning rates, discount horizons, and batch sizes, identifying the globally optimal hyperparameter configuration.
Modeling inter-crop biological compatibility under fluctuating climate and soil conditions is traditionally hampered by isolated agronomic assumptions that fail to generalize across millions of crop-pair permutations.
Formulated four novel algorithmic components: Nutrient Balance Index (NBI), Environmental Resilience Score (ERS), Crop Compatibility Index (CCI) with logistic conflict dynamics, and the unified AgriCompatNet ML pipeline; trained and validated on 2,200 crop instances and evaluated across 2.4 million generated crop-pair combinations.
Hyper-realistic facial reenactment and deepfake video manipulation present severe identity spoofing and security risks that evade conventional CNN-based detectors lacking long-range spatial attention.
Engineered a frame-level deepfake video detection pipeline leveraging a Vision Transformer (`vit_base_patch16_224`) with ImageNet pretrained weights; implemented automated 1 fps temporal frame extraction across original and manipulated video sequences (DFD datasets), analyzing patch attention weights to classify frames as real or manipulated.
An active intellectual property portfolio comprising 5 published Indian patents and 3 pending patent applications, backed by industry cloud certifications and peer-reviewed technical publications.
An edge-optimized smart irrigation architecture incorporating distributed Raspberry Pi units, multi-channel sensing, and low-latency micro-controller execution for autonomous precision water delivery.
A machine learning and rule-augmented plant irrigation system dynamically adjusting watering intervals based on real-time soil moisture gradient feedback and environmental forecasts.
Methodology and execution pipeline governing automated valve sequencing, hydraulic flow balance, and failsafe cloud override logic across large-scale agricultural plots.
AI-driven controller mechanism leveraging temporal forecasting models to continuously optimize moisture thresholds and reduce water consumption in agricultural zones.
Affective computing system capturing multimodal emotional metrics to formulate personalized self-care interventions and stress alleviation routines.
An automated track monitoring system employing edge sensors, high-speed acoustic diagnostics, and GPS fault-localization algorithms for continuous railway safety inspection.
Engineered during the Smart India Hackathon (SIH 2025) National Finals (Top 30 Hardware Teams nationwide for KSEB). Features LoRa edge decision nodes, Kafka data streams, and encrypted power grid phase balancing.
A wearable smart spectacle attachment featuring micro-compartment storage, bluetooth app synchronization, and auditory-visual medication schedule compliance reminders.
Rigorous foundations in Computer Science & Engineering, Artificial Intelligence, and Applied Mathematics with a sustained record of top academic distinction and engineering leadership.
Bachelor of Technology in Computer Science and Engineering (Specialization in AI & ML)
Higher Secondary Certificate (Class XI–XII, Science & Engineering Stream)
Secondary School Examination (Class X CBSE Board)
I am actively exploring AI Engineer, Machine Learning Engineer, Software Developer, Backend / API Engineer, Cloud Engineer, and Forward Deployed Engineer (FDE) opportunities globally.