SachinJain

Interview preparation is usually fragmented across notes, documents, and generic chat threads. That makes answers harder to refine and harder to reuse.

Users work through answers, save stronger material, and build a structured base of moments and playbooks that can be reused across interview scenarios.

Designed system architecture across backend, frontend, Data, and AI layers with a multi-step workflows.Implemented context-aware retrieval for personalization.

The product became a good test of how AI feels when it is embedded into a workflow instead of acting like a single detached chat box. Contract-based communication improves reliability between agents

The challenge was to create something that organize routes, timing, and stops into a single flow but leave room for personalization.

The system takes trip intent and transforms it into structured, multi-day planning with more useful pacing, organization, and itinerary detail.

Defined a structured output format for itineraries (routes, days, stops, stays). Iterated on prompt design to generate consistent. Focused on sequencing and flow rather than isolated recommendations

AI outputs improve when the prompt aligns with intent rather than over-specifying behavior. Too much constraint reduces usefulness, while the right level of openness improves outcomes.

Research material is scattered across multiple papers and it is time-consuming to extract and connect key insights.

Documents are processed and stored in a FAISS vector database. Relevant sections are identified based on the user query. Retrieved content is processed to generate clear, grounded outputs

Built a custom RAG pipeline and indexed a FAISS-based vector database for efficient retrieval. Implemented a two-pass approach to extract relevant context along with source references for verification.

Chunking strategy and token usage must be balanced — too little context reduces quality, while too much increases cost without better results.