I design and evaluate evidence-grounded RAG pipelines, in-process analytical engines with DuckDB, and robust data platforms — backed by production full-stack engineering to take systems to real users.
Resume Roaster — Founder & Developer (2026 – Present)
An AI-powered ATS resume optimizer and career preparation studio built with Next.js, TypeScript, Supabase, and the OpenAI API. Features a free ATS scanner tested against major enterprise platforms (Workday, Greenhouse, Taleo, iCIMS, Lever), an automated bullet rewriter, a cover letter generator, an AI mock-interview simulator, and PDF export via @react-pdf/renderer. Shipped and live for paying users.
| Project | Description | Primary Stack |
|---|---|---|
| Agentic RAG Knowledge Assistant | Autonomous RAG system with query decomposition, hybrid lexical/vector search (BM25 + NumPy RRF |
Python · OpenAI API · Rank-BM25 · Pydantic · Streamlit |
| Turbine AvgEGT Regression | Industrial turbine engine Exhaust Gas Temperature (AvgEGT) regression for predictive maintenance ( |
Python · XGBoost · Scikit-learn · Pandas · Telemetry Analytics |
| Customer Churn & Retention Modeling | End-to-end subscriber churn risk modeling on 7,043 customer accounts using SMOTE and Random Forest (ROC-AUC: 0.823) with standalone CLI inference. | Python · Scikit-learn · Imbalanced-Learn · Random Forest · Seaborn |
| AI Data Analyst Agent | Autonomous tabular analytics agent with in-process DuckDB columnar OLAP, AST-based read-only SQL validation (33 disallowed keywords), and numerical faithfulness verification. | Python · DuckDB · Plotly · Streamlit · Pandas |
| Erasmus Mundus Scholarship Tracker | Autonomous monitoring agent tracking European scholarship deadlines (EMAI, EDISS, DEAI) with resilient DOM scraping, state caching, and automated Telegram dispatch. | Python · BeautifulSoup · Telegram Bot API · Docker · Automation |
| Jan Darshan | Independent civic and legislative intelligence platform tracking Indian MPs and ministers with automated scraping, data normalization, and interactive constituency mapping. Live at jan-darshan.vercel.app. | Next.js · TypeScript · MongoDB · Puppeteer · Upstash Redis |
| GSTify | GST-compliant invoicing web application for Indian freelancers with automated tax computation, PDF invoice generation, and serverless persistence. | Next.js · TypeScript · Drizzle ORM · Neon PostgreSQL · pdf-lib |
| SubsSync | SaaS recurring-billing platform for Indian businesses featuring UPI AutoPay, automated dunning cycles, and transactional WhatsApp notifications. | Next.js · Razorpay · UPI AutoPay · WhatsApp API |
| Aarohana Trails | High-concurrency trekking and adventure expedition booking platform featuring an atomic seat ledger, custom headless CMS, dynamic itineraries, and real-time operations audio alerts. Live at aarohanatrails.com. | Next.js 16 · TypeScript · PostgreSQL · Prisma ORM · Tailwind CSS |
| Namma Metro Guide | Android transit navigation guide for Bengaluru Namma Metro with route planning, fare estimation, and interchange mapping. | Java · Kotlin · Android Studio |
- Deterministic Guardrails First: I prioritize architectural guardrails (read-only SQL AST filtering, strict Pydantic validation schemas, and post-generation citation checking) rather than relying on unconstrained LLM code execution.
- Measurable & Evaluated: All AI and retrieval systems are rigorously benchmarked against baseline implementations with automated test suites and metric evaluation (faithfulness, precision, and latency).
- Real-World Business Grounding: Before focusing full-time on software and data systems, I owned and operated an independent restaurant in Varanasi — managing daily operations, vendor contracts, cash flow, and P&L. This operational background directly informs how I build software: grounded in real business constraints, strict unit economics, and operational reliability.
- Data & AI Core: Python · SQL · DuckDB · Pandas · NumPy · Scikit-learn · XGBoost
- AI & LLM Systems: Retrieval-Augmented Generation (RAG) · OpenAI API · BM25 · Pydantic · Prompt & Output Evaluation
- Databases & Storage: PostgreSQL · Supabase · MongoDB · Redis · REST APIs · Firebase
- Supporting Full-Stack: TypeScript · JavaScript · Next.js · React · Node.js · Streamlit · Tailwind CSS
- Tools & Analytics: Git · GitHub · Plotly · Power BI · Matplotlib · Seaborn
- Mobile & Platforms: Java · Kotlin · Android Studio · Vercel
I write in-depth engineering breakdowns on data architectures and LLM systems at rajchhapariya.space/blog:
- Hybrid Retrieval Systems in Production: Combining BM25, Dense Embeddings, and Reciprocal Rank Fusion (RRF).
- Deterministic Guardrails for SQL-Generating LLM Agents: From Natural Language to Verified AST Execution.
- In-Process Columnar OLAP with DuckDB Architecture: Replacing Heavyweight Data Warehouses for Analytical Microservices.
- Evaluating Hallucination & Citation Faithfulness in RAG: Methodologies for Context Grounding and Atomic Claim Recall.
- Next.js App Router Architecture: Server Components, Streaming SSR, and Static Site Generation.
- 🌐 Portfolio & Case Studies: rajchhapariya.space
- 💼 LinkedIn: linkedin.com/in/rajchhapariya
- 𝕏 X: x.com/raj_chhapariya
- 🐙 GitHub: github.com/Rajchhapariya
- 📧 Email: rajchhapariya8@gmail.com