Final-year CSE AI undergrad, spending most of my time on LedgerMind — an agentic financial operating system that acts as a virtual accountant, bookkeeper, analyst, and CFO for small businesses.
The idea I keep coming back to: a chain of AI agents can read a receipt, book the expense, flag the fraud, and watch it against a budget — but every number has to be computed deterministically. The LLM reasons and explains. It never invents a figure.
|
Agentic finance OS for small businesses — receipt OCR → categorization → GST/ITC intelligence → fraud checks → budget alerts, with a tool-calling CFO chat over your own books.
|
Multi-agent banking assistant for NepGlish (Nepali–English code-mixed speech). A fine-tuned 8B SLM handles sub-second intent classification locally, replacing cloud LLM calls.
|
|
Audits software requirement specs for ambiguity, then synthesizes editable Use Case / Activity / DFD diagrams straight from natural language.
|
🗂️ ClarityStackLocal-first knowledge system that turns long AI chats into structured, version-controlled knowledge — while keeping raw history as immutable ground truth.
|
🖥️ Nishant OSPortfolio where the website is the project — canvas neural-net background, interactive knowledge graph, motion-first throughout.
|
Computer vision, domain adaptation, style transfer, maritime risk intelligence, ROS2 robotics — the rest of the workbench. |
AI / ML
Backend
Frontend
