AI/ML Engineer | Full-Stack Developer
I build production ML systems and full-stack prototypes that work at scale. Focused on agentic AI, LLM systems, and rapid-iteration development.
Currently: USAR (GGSIPU), 3rd Year AI/ML
Exploring: Distributed inference, fine-tuning strategies, systems design for ML, Token Optimization
Waste management lacks efficient automated visual sorting. Industrial buyers can’t connect with scattered recyclable collectors at scale.
Built an edge AI system that classifies waste from images with 97%+ precision. Runs on low-end hardware (GTX1650 + 4GB RAM), enabling deployment at collection points.
Bridges waste collectors ↔ industrial buyers. Proves scalable synthesis engine for waste-to-resource workflows.
The original university student portal (Examweb GGSIPU) suffered from severe UX issues, slow loading times, and a lack of integrated tools for academic tracking.
Developed a comprehensive, React-based enhanced portal that aggregates results, attendance, and resources into a single intuitive dashboard. Engineered a custom marks calculator and optimized the data fetching pipeline to ensure the site operates with zero noticeable latency.
Reduced page load time from ~3-5s to instant (0-200ms TTFB). Used by 200+ students for academic tracking.
Long, verbose conversational prompts rapidly inflate Large Language Model (LLM) API costs and consume context windows unnecessarily.
Engineered a specialized Regex pipeline + prompt optimization technique that scrubs redundant tokens while preserving semantic meaning. Strips structural bloat while mathematically preserving core intent.
Reduced token consumption by ~10% per prompt, cascading into 40% total API cost savings across prolonged sessions.
For teams running LLM applications at scale, prompt efficiency directly impacts unit economics. This proves optimization-first thinking for production systems.
Individuals with visual, auditory, or cognitive disabilities face fragmented assistive tools with high latency and poor real-time spatial support.
Led cross-functional team to build an all-in-one Flutter accessibility suite. Integrated Google ML Kit for real-time on-device object detection, AR spatial navigation, and OpenAI Whisper for zero-latency speech synthesis.
Delivered unified, low-latency assistive app enabling real-time spatial and speech recognition without cloud dependencies.
Existing personal safety apps rely on reactive emergency contacts rather than preventative spatial intelligence and hardware-linked trigger response.
Architected a preventative safety app in Flutter & Figma featuring live risk-zone heatmaps, hardware-linked scream detection for automated SOS triggering, and verified community updates.
Designed end-to-end UX/UI and technical blueprint for a proactive, hardware-connected safety ecosystem.
Evaluating ground-level implementation of public policy relies on slow, bureaucratic surveys prone to data abstraction and reporting bias.
Built a data pipeline and predictive AI engine processing 15GB+ of raw Indian government datasets. Correlated regional expenditure, demographic metrics, and execution logs to score policy efficacy.
Achieved 98%+ prediction accuracy in estimating regional policy implementation quality, creating a scalable blueprint for algorithmic civic governance.
Students waste significant time manually organizing fragmented study resources and tracking academic backlogs across disparate platforms.
Engineered an AI study assistant powered by batched Gemma LLM inference. Automatically parses syllabus patterns, curates structured video playlists, and builds personalized study roadmaps.
Reduced LLM API overhead via batching while providing an adaptive, automated study planner that eliminates manual course organization.
Smallholder farmers lack hyper-local, offline crop advisory tools, forcing reliance on expensive cloud services or outdated regional defaults.
Built an offline-first React + ML client that executes predictive agronomy models directly on-device. Analyzes local soil and weather metrics to yield custom crop recommendations without cloud connectivity.
Achieved 85–90% forecasting accuracy for crop yield and planting timing on low-spec edge hardware with zero internet dependency.
Domestic water wastage goes unnoticed until high utility bills arrive due to lack of granular usage analytics.
Developed an agentic chatbot (IBM CSRBox Internship) leveraging Gemini API. Parses user utility bills via vision/multimodal input, calculates consumption trends, and generates actionable water reduction strategies.
Automated utility bill analysis and predictive consumption forecasting, delivering tailored reduction plans directly to consumers.
How to use
Legend
Python • NumPy • Pandas • Scikit-learn • TensorFlow • OpenCV • NLP • LLMs • RAG
FastAPI • Django • Node.js • Express • PostgreSQL • MongoDB • Docker • Linux
React.js • Next.js • TypeScript • Tailwind CSS • HTML/CSS • Flutter • Dart
Git • GitHub • VS Code • Figma • Canva • Prompt Engineering • Cursor
Proficiency: Advanced in Python, React, FastAPI, and ML fundamentals. Intermediate in mobile development (Flutter), backend systems design, and AI tooling.