I specialize in:

• Software Engineering
• Web Development
• Graphic Design
• Problem Solving
• Creative Thinking

Currently working as Co-Founder and Software Engineer at Navdyut AI Tech and Research Labs Pvt. Ltd., where I lead AI research and develop innovative solutions.

Tech Stack: React, TypeScript

Status: active

Repository: https://sonification.shiva.codes

Responsibilities:

Develop websites and systems to be used by its clients and maintain current existing websites and systems.

Technologies Used: ReactJS, NextJS, Frame Motion, ThreeJS

Responsibilities:

Built internal ERP modules for billing, vendor, and project tracking used by construction operations. Developed semantic project document search using RAG pipelines with OpenAI embeddings and vector databases for construction drawings, BOQs, and reports. Optimized API performance and database queries improving internal tool response times and reliability.

Technologies Used: ReactJS, NextJS, Tailwindcss, Prisma, Supabase, OpenAI, Vector DB, RAG

Responsibilities:

Built an Assamese chatbot on a 22B Mistral model with RAG pipelines for translation and government applications. Scaled the system to 500+ users and contributed to deployments for public sector use. Project work was featured in regional newspapers.

Technologies Used: Mistral, RAG, Langchain, LlamaIndex, Pinecone, NextJS, Tailwindcss, Supabase

Responsibilities:

Reduced jewelry design turnaround from days to minutes by building AI agent workflows using the Vercel AI SDK with observability via Langfuse. Built an interactive jewelry design canvas using React + XYFlow enabling credit-gated editing and real-time agent-driven design iteration. Developed a Redis queue pipeline generating 70+ jewelry design variations in 5 minutes per batch. Built a custom memory system with rolling context per user to retain design preferences and reduce duplicate generations. Improved reliability and performance across microservices deployed with Docker, Firebase, and GCP.

Technologies Used: Vercel AI SDK, React, XYFlow, Redis, Firebase, GCP, Docker, Langfuse, NextJS

Abstract:

A non-autoregressive text-to-speech (TTS) multilingual synthesis framework designed to address the linguistic diversity and real-time deployment challenges of Indian languages. By deploying a unified encoder-decoder architecture paired with lightweight hidden-state adapters, PolySpeech-HS enables efficient cross-lingual generalization while preserving language-specific prosodic nuances. Achieved state-of-the-art performance with MOS of 4.30, MCD of 4.7 dB, and RTF of 0.13 across six Indian languages.

Journal: IEEE Transactions on Audio, Speech and Language Processing

Year: 2025

Collaboration: Vellore Institute of Technology

Status: Under Review

Technologies/Methods: TTS, Non-Autoregressive, Hidden-State Adapters, Multilingual AI, Indian Languages, AMO-HSA

Abstract:

A data-centric, hardware-light workflow for fine-tuning transformers that sidesteps costly LLM APIs. Automatically scrapes high-signal web content and converts it into Q&A pairs to fine-tune a GPT-2-Medium model (355M parameters) in ~7 minutes on a single RTX-3060. Achieves 67.3% accuracy (+34% over base model) with 1.4s median latency and zero inference cost.

Journal: IEEE Transactions on Computational Social Systems

Year: 2025

Collaboration: Vellore Institute of Technology

Status: Under Review

Technologies/Methods: GPT-2, LoRA, 8-bit Adam, Domain Adaptation, Next.js, Q&A Generation, Fine-tuning

Abstract:

The first fine-tuned Large Language Model specifically engineered for Assamese, a low-resource Indo-Aryan language spoken by approximately 15 million individuals. Introduces AssamText-750K dataset and custom Unicode mapping system exclusively for Assamese. This pioneering work becomes the first and only Assamese LLM backed by language-specific Unicode infrastructure, achieving 20% average improvement across text generation fluency, sentiment analysis accuracy, and Assamese-to-English translation quality.

Journal: IEEE Transactions on Neural Networks and Learning Systems

Year: 2025

Collaboration: Vellore Institute of Technology

Status: Under Review

Technologies/Methods: Mistral 22B, LoRA, Unicode Mapping, Assamese NLP, Low-Resource Languages, AssamText-750K

You can reach out to me through my portfolio website for any opportunities, collaborations, or just to say hello.

Let's build something amazing together! 🚀