Hi, I'm Harshal Manerikar

Problem

Extracting insights from large document collections requires expensive custom engineering or poor off-the-shelf tools that don't handle mixed formats or nuanced Q&A.

Solution

End-to-end RAG (Retrieval-Augmented Generation) platform built with FastAPI and React. Users upload documents and ask natural-language questions answered by OpenAI or Gemini.

Problem

Warehouses operating in isolation hold either too much or too little stock, driving up costs and creating shortage risk. Manual rebalancing is slow, error-prone, and unable to react to changing demand.

Solution

Predictive inventory optimization platform that mines warehouse and order data to forecast demand, calculate inventory pressure, and recommend optimal stock transfers between locations.

Problem

Crypto trader performance shifts dramatically with market sentiment, but the patterns behind those shifts are obscured by noisy, high-volume trade data and lack of structured behavioural segmentation.

Solution

Data-driven analysis pipeline merging Hyperliquid trader data with the Bitcoin Fear & Greed Index to surface how sentiment regimes affect win rates, risk-taking, and trader behaviour.

Problem

Traditional social platforms centralise all user data on company servers, creating massive privacy risks and making regulatory compliance (GDPR, HIPAA) nearly impossible.

Solution

Federated learning architecture where ML models train locally on each device. Only encrypted gradient updates are shared with the server; raw user data never leaves the client.