B.Tech AIML student at Walchand College of Engineering, Sangli. Creator of Zenis (Open-Source AI Presentation Engine), eduWin (B2B SaaS Coaching OS), and Parakh (Consumer Health Startup). Focused on deterministic AI, strict unit economics (<$0.005/run), and production software.
Three distinct engineering endeavors: an open-source AI engine, a deployed B2B SaaS platform, and a consumer startup app.
Ultra-low-token AI presentation generator. Replaces expensive frontier prompts with deterministic scrapers and single-forward-pass System-1 gating to cut token costs by 90%.
Turn-key operations and financial management blueprint for coaching institutes. Enforces strict tenant isolation with Postgres Row-Level Security and serverless cloud deployment.
Packaged-product intelligence platform. Scans barcodes and nutrition labels, validating claims against statutory FSSAI & ICMR-NIN thresholds via Gemini 2.0 Flash VLM.
Complete stack across language runtimes, machine learning pipelines, backend services, databases, and client interfaces.
Interactive architecture simulators directly showcasing my three main projects.
Frontier models re-ingest entire web crawls for every slide. Zenis runs single-pass System-1 filtering and in-memory greedy layout reduction, slashing input fluff by 90% before calling the synthesis model.
Offloads 90% of processing from LLMs to open-source scrapers (Trafilatura, PyMuPDF) and lightweight decision models. Uses a two-pass slide reducer and micro-synthesis to hit <$0.005 cost per deck with sub-8s generation.
SaaS platform for tuition centers covering batch timetables, attendance, and fee tracking. Uses Postgres Row-Level Security for strict tenant isolation and rotating JWT refresh tokens.
Mobile and web app that evaluates food, beverage, and cosmetic labels against statutory FSSAI and ICMR-NIN limits using barcode scans, package photos, or quick-commerce links.
Interactive docs learning platform with line-by-line code explanations using DeepSeek R1, ChromaDB, and FastAPI SSE streaming.
Android app for student campus living with roommate matching, expense splitting, and housing listings built with Kotlin and Jetpack Compose.
Scores 100k candidate profiles in 1.00s using memory-mapped embeddings and anomaly penalties.
Extracts legal clauses, assesses risks, and explains contracts in plain English using Gemini Flash.
Retention model predicting drop-off from early webinar activity; segmented users into 4 action tiers.
Sales demand forecasting with Isolation Forest anomaly detection and K-Means demand clustering.
Campus navigation engine built on OpenStreetMap data with side-by-side search visualizer.
Built time series forecasting models using SARIMA, Meta Prophet, and XGBoost to optimize sales predictions across multi-year retail transactions. Segmented product categories by demand volatility with K-Means.
Analyzed webinar engagement data using Python and SQL. Built a tuned logistic regression churn model achieving 0.9806 ROC-AUC across 19,889 user sessions, segmenting into 4 retention action tiers.
Built and deployed responsive web applications using React, Flask, and modern CSS.
Coordinated student engineering initiatives, bridging student body operations and administration to drive campus technical events.