DHANUSH ARAVIND A

Built a LangGraph self-correcting AI agent that connects to multiple Databricks Managed MCP servers (Genie for natural language to SQL, Vector Search for semantic retrieval) to generate detailed, strategic business reports. The agent plans its retrieval, evaluates whether the gathered evidence is sufficient, and reformulates and retries when grounding is weak, producing answers with every claim cited to a specific source and no hallucination. Underneath, a Medallion lakehouse on Databricks (Bronze, Silver, Gold) processes synthetic CRM, ERP, and support data. The agent never queries the database directly, it reasons only over governed, read-only MCP endpoints, and delivers grounded strategic insights through a Streamlit interface.

• Langgraph
• SQL
• Pyspark
• Open Ai api
• DataBricks
• Genie
• Vector search
• Medallion architecture
• Excalidraw mcp
• streamlit

Designed a 3-layer architecture consisting of a PostgreSQL warehouse with analytical views, a FastAPI-based MCP layer exposing governed REST endpoints, and an AI reasoning layer that ensures the LLM never queries the database directly. Built Python ETL pipelines to ingest SaaS business data into structured warehouse views covering MRR, churn by plan, revenue by region, and support-churn correlation. Deployed the backend on AWS EC2 with RDS and S3, incorporating session-based conversation memory, environment-isolated credentials, and a lightweight chat UI for natural language business querying.

• Python
• FastAPI
• PostgreSQL
• MCP
• OpenAI
• AWS

Built a lightweight AI-powered web app that roasts your resume, upload a PDF, extract the text, and receive a brutally honest AI critique rendered in Markdown. Engineered with a clean modular structure separating PDF extraction, LLM prompting, and Streamlit UI into dedicated components. It lets you improve your resume by getting brutally honest advice.

• Python
• pdfplumber
• OpenAI API
• Streamlit

Designed a batch ETL pipeline using Medallion Architecture (Bronze, Silver, Gold layers) to transform raw sales CRM and ERP data into analytics-ready outputs, cleaning and transforming over 100K sales records using SQL. Deployed a containerised SQL Server instance via Docker, integrated with SSMS for structured data validation and downstream querying.

• SQL
• Docker
• CRM Data
• ERP Data
• SSMS

Analysed global salary and purchasing power parity data to identify where entry-level roles deliver the highest real income value. Cleaned and processed 1M+ rows across 160+ columns, then built an interactive 3D world map in Power BI enabling visual comparison of purchasing power across countries making it easy to pinpoint where your salary goes furthest.

• Power BI
• Python
• Pandas

Analysed 5,000+ customer reviews to classify sentiment and uncover recurring pain points and strengths shaping brand perception. Improved sentiment analysis accuracy by 18% through feature selection and model tuning, and visualised key themes and trends via dashboards to support stakeholder decision-making.

• Sentimental Analysis
• Python
• Power BI
• Pandas
• Skikit-learn
• NLP
• Feature Selection

Fine-tuned a pre-trained text-to-image diffusion model using LoRA (Low-Rank Adaptation) and PyTorch to generate accurate yoga poses from natural language prompts. Curated and preprocessed a domain-specific image-caption dataset, handling pose alignment, caption normalisation, and training splits. Trained and evaluated LoRA adapters on GPU-based pipelines, tracking output consistency and prompt adherence, with the full training and inference workflow containerised in Docker for reproducibility across Linux environments.

• Python
• PyTorch
• Diffusion Models
• LoRA
• Docker

Built a multilingual AI assistant using OpenAI API and FastAPI that accepts text or voice input, explaining bank alerts, government notices, and WhatsApp messages in simple Tamil, Tanglish, or basic English, designed for users with limited digital literacy. Implemented scam detection logic to flag messages impersonating official institutions, returning severity assessments and safe suggested replies to reduce user risk. Integrated Speech-to-Text and Text-to-Speech pipelines to enable voice-first interaction for users with limited smartphone familiarity.

• Python
• FastAPI
• OpenAI API
• TTS
• STT

Built a full-stack portfolio builder that lets anyone create and publish a professional portfolio for free within seconds. Features a multi-step creation flow covering personal info, skills, projects, and template selection, with three distinct portfolio templates and file upload support for profile photos, project images, and resumes. Includes a public Explore directory with search, skill filters, availability filtering, sort, and pagination, alongside QR/link sharing and view tracking for portfolio visibility and deployed as exact website we are using now.

• Typescript
• HTML
• CSS
• Next.js App Router
• Postgres

AI-powered job market analytics platform that analyzes Kaggle data to forecast salary trends, benchmark highest-paid roles, and map in-demand skills across 12+ data science job categories. Features interactive visualizations for salary predictions (2020-2028), work preference analytics, and real-time tools/skills requirement mapping. Built with Kaggle and responsive dashboard UI.

- Built an AI-powered IT service desk assistant using LangGraph to automate common Level 1 support requests, reducing manual triage effort through intent-based workflow orchestration. - Developed an intent classification pipeline supporting password resets, Microsoft 365 licensing, Outlook, Teams, and device health requests, achieving over 90% intent classification accuracy during testing. - Integrated Microsoft Graph API to automate user account lookups, group membership management, license assignment workflows, and directory information retrieval. - Automated IT support workflows using Power Automate, reducing repetitive service desk actions and enabling end-to-end execution for common user requests. - Integrated Jira Service Management APIs to automatically create, update, assign, and e