Selected Works.

A curated deep-dive into production-grade systems, RAG architectures, and multi-agent workflows. Designed for scale, compliance, and performance.

01 — GenAI

FinSmartAI

Full-stack GenAI platform for financial document Q&A and market sentiment analysis.

Source

The Problem

Financial analysts waste hours manually parsing complex compliance documents, and standard LLMs hallucinate critical financial figures.

The Solution

Designed an end-to-end RAG pipeline using AstraDB vector search that grounds LLM responses exclusively in cited financial documents, alongside a real-time market sentiment module.

Architecture

FastAPI backend handles document ingestion and semantic search. React frontend delivers a seamless chat interface. LangChain orchestrates the retrieval augmented generation.

Impact & Metrics

Achieved BERTScore F1 0.8605 and Semantic Similarity 0.834, significantly reducing hallucinations. Published peer-reviewed research in IJESAT (Vol. 26, Issue 4, Apr 2026).

Key Learnings

Learned how to rigorously evaluate RAG pipelines using BERTScore and manage large-scale vector embeddings in a production environment.

PythonFastAPILLMsRAGAstraDBLangChainReact

02 — GenAI

Audit IQ

Autonomous multi-agent system for automated compliance validation.

Source

The Problem

Manual compliance auditing is slow, error-prone, and struggles to scale across thousands of highly dense legal and financial documents.

The Solution

Built an autonomous multi-agent orchestration system using CrewAI. Specialized agents handle OCR extraction, rule-based validation, and dynamic risk scoring.

Architecture

PyMuPDF and Retriever OCR extract context. CrewAI delegates tasks to specialized LLM agents. Pinecone acts as the memory bank for regulatory context.

Impact & Metrics

Reduced audit time by ~60%. Secured a National Finalist position at the Google Agentathon (36-hour hackathon).

Key Learnings

Mastered multi-agent orchestration patterns and handling non-deterministic LLM outputs in strict, rule-based compliance environments.

CrewAILangChainFastAPIPineconeReactPython

03 — CV

AI Identity Verification & Fraud Detection

Multimodal OCR and Machine Learning pipeline for KYC automation.

Source

The Problem

Traditional KYC processes require intensive manual review and are susceptible to sophisticated document forgery.

The Solution

Architected a multimodal pipeline that extracts text via OCR and applies ensemble ML classifiers to detect anomalies and fraudulent patterns in identity documents.

Architecture

Flask REST APIs expose the inference engine. The entire service is containerized via Docker for horizontal scaling, with real-time monitoring dashboards.

Impact & Metrics

Improved automated KYC verification accuracy by ~35% and cut manual review time during the Infosys Springboard internship.

Key Learnings

Gained enterprise-level experience in containerizing ML inference APIs and optimizing for sub-second response times.

PythonComputer VisionOCRScikit-learnFlaskDocker

Other Engineering Work.

UIVerse

GenAI

Multi-agent runtime workspace generator where AI dynamically generates interfaces in real-time. Built at AI Tinkerers Global Hackathon.

ReactCopilotKitCrewAIGemini 2.5 Flash

GenAI Career & Skills Advisor

GenAI

Recommendation system for career pathways mapping user profiles to skill gaps using prompt engineering.

PythonGeminiLLMs

OpenClaw Desktop Assistant

Automation

Locally running desktop automation assistant for seamless workflow management.

PythonAutomationOS APIs

Plant Disease Detector

Computer Vision model designed to identify crop diseases from leaf images.

PyTorchComputer VisionPython

House Price Prediction

End-to-end machine learning pipeline utilizing regression models to forecast real estate prices.

Scikit-learnPandasFlask

Internship EDA Suite

Data Science

Collection of advanced Exploratory Data Analysis notebooks and visualizations created during Edunet Foundation internship.

PandasNumPyMatplotlib