Doctoral research at NIT Goa on hardware acceleration of Generative Adversarial Networks, alongside applied cybersecurity and intelligent-systems work. Each area below is backed by publications and, where it exists, open-source implementation.
Resource-efficient synthetic image generation: multifractal and evolutionary methods embedded in GAN training so models are small, stable and deployable on constrained edge hardware.
FPGA / VLSI design for AI: synthesising GAN discriminators on the AMD Kintex-7 KC705 and earlier Vedic-math arithmetic cores, targeting low latency and power at the edge.
Applied security through teaching: Incident Response and Vulnerability Assessment & Penetration Testing at the National Forensic Sciences University, plus forensic and malware-research experiments in the open.
Optimisation-driven learning: PSO and hybrid evolutionary techniques for GAN training, and exponential-variation PSO for analog circuit sizing in constrained design spaces.
Research and software are treated as one pipeline. Every paper maps to a method, an implementation and — where it exists — an exact source repository. Nothing here links to a vague "view profile".
Some publications have no public code yet — those are listed as paper-only rather than inventing links. Verified implementations are shown with their exact repository.
Every entry verified against ORCID and Crossref. DOIs resolve to the publisher; the two conference papers without DOIs link to their ORCID records.
The pinned, maintained core of the charudatta10 profile. Every link resolves to the exact repository or site.
A ~160-line minimalist LLM framework for Lua. Zero dependencies, zero vendor lock-in; agents, workflows and RAG in one file.
Static site with a client-only RAG chatbot over the founding figures of forensic science — WebLLM in the browser, no server, no API keys.
A composable, SPDX-compatible licensing framework — modular legal clauses instead of rewriting licenses, with monetisation built in.
Aggregated link hub — a curated, categorised collection of useful resources with custom link icons and fast setup.
Automation, AI tooling, security and knowledge systems built for research workflows and teaching.
Automates init, deploy, licence, readme and maintenance tasks across projects (now maintained as PowerShell modules).
Agent framework, retrieval-augmented generation and code translation, each running on local models.
Generate project documentation with a local LLM, produce illustrations locally, and a lightweight local chat interface.
Secure single-download file sharing, and a Lua reverse-engineering sandbox with static and dynamic analysis.
Foam + Raito personal knowledge graph, and a Docsify-powered notes & wiki site.
Showcase projects as tiles, ship any Python script as a single-file PWA, and organise files by tags.
Code directly tied to the publication record — the same lines of work from the earliest papers to the doctoral GAN research.
Handwritten Devanagari script recognition system from the Bachelor's project (2015) — the same line of work as the ISVLSI publication on FPGA pre-processing.
2019 GAN research implementation (perceptual adversarial training experiments) — part of the doctoral exploration into stable GAN training that led to the INDICON publication.
Degree projects that started the research lines now published.
Exploratory and laboratory work — some published, some living on the profile, all listed without invented links.