Anish Manchanda

Designed and shipped a production RAG pipeline for CUET exam preparation, combining FAISS vector search with BM25 lexical ranking in a hybrid dual-stream retrieval architecture using HuggingFace Sentence Transformers. Implemented citation-aware context construction and a dedicated Previous Year Question (PYQ) retrieval stream to measurably reduce LLM hallucinations and improve answer traceability for end users. Integrated and migrated LLM backends (Mistral → Google Gemini 2.5 Flash), engineering custom system prompts and few-shot prompt templates to enforce exam-focused, grounded responses.

AI/ML Intern - Pepper Labs AI

(2026-02 - 2026-04)

• Designed and shipped a production RAG pipeline for CUET exam preparation, combining FAISS vector search with BM25 lexical ranking in a hybrid dual-stream retrieval architecture using HuggingFace Sentence Transformers.
• Implemented citation-aware context construction and a dedicated Previous Year Question (PYQ) retrieval stream to measurably reduce LLM hallucinations and improve answer traceability for end users.
• Integrated and migrated LLM backends (Mistral → Google Gemini 2.5 Flash), engineering custom system prompts and few-shot prompt templates to enforce exam-focused, grounded responses.
• Diagnosed and resolved a critical triple-invocation concurrency bug in the LLM call chain; hardened the pipeline with robust error handling and end-to-end Jupyter Notebook documentation for seamless technical handoff.

B.Tech - Computer Science (Artificial Intelligence) - University School of Information Communication & Technology (USICT) (2024 - 2028)

Senior Secondary (Class 12) - Bal Bharati Public School

Secondary (Class 10) - Bal Bharati Public School