Open to opportunities

Sanskar Narlawar

@sanskarnarlawar

Message

AI-focused Founding Engineer building production agent systems and backend platforms for enterprise impact.

India

Message

What I'm looking for

I want to build reliable AI agent systems and backend platforms that turn large, messy data into secure, user-friendly products—shipping measurable improvements, collaborating closely, and continuously validating models before deployment.

I’m a Founding Engineer building production AI agent systems at Merlin AI (YC S24), where I design agents for bulk data upload, custom report generation, and validated query execution. I use reason–act loops with tool calling, a result-handle architecture for conversational report insight, and execution-based evaluations to catch regressions before deployment.

Previously at Deutsche Bank, I built an internal RAG chatbot that reduced manual effort by 30%, with indexed retrieval, user-specific access controls, and session-based conversation memory for 100+ active conversations. I also worked on reconciliation ML (fuzzy matching at 50k+ ledger entries) and delivered backend microservices in Java 8/Spring Boot—improving database load by 30% through batch processing and modernizing legacy integrations.

Experience

Work history, roles, and key accomplishments

Current

Founding Engineer

Current

Merlin AI

Oct 2025 - Present (9 months)

Built production AI agents for bulk data upload and custom report generation, including interactable report conversations backed by validated MongoDB aggregations. Led core backend work including OpenSearch-based search, Debezium CDC to Kafka, and a Materials module with layered pricing and forecasting.

Agentic AI Tool calling TTL Collections Result Handle Architecture OpenSearch Elasticsearch Style Indexing Debezium Kafka Pinecone Vector Databases Material Pricing Forecasting

Software Developer

Deutsche Bank

Jul 2023 - Sep 2025 (2 years 2 months)

Developed an internal RAG chatbot to answer application queries with embedded Confluence retrieval, secure user-specific access controls, and session-based conversation memory. Built supporting ML and microservices, including reconciliation using fuzzy matching and a Java/Spring Boot microservice architecture that routes requests across databases and integrates with message queues.

RAG Embeddings Vector Search SQL Confluence Indexing Access Control Citations Session IDs EDA Fuzzy Matching Microservices Java 8 Spring Boot Spring Data JPA Apache Camel Message Queues Batch Processing

Lovable For ERP

Together Fund Hackathon

Jul 2025 - Present (1 year)

Built an invoice-to-ERP workflow using document parsing and text-to-speech, integrated with WhatsApp. Exposed ERPNext functions as agent tools using Google ADK to populate and create a customized ERP.

Text To Speech WhatsApp Integration ERPNext Google ADK Function Whitelisting

Matrix and Primes Library

Summer of Codes

Enhanced an open-source C++ scientific computing library similar to NumPy/MATLAB by adding modular arithmetic support. Implemented a ModInt library using the Chinese Remainder Theorem with Garner’s Algorithm and validated it with CMake/GoogleTest.

C++Modular Arithmetic ModInt Library Chinese Remainder Theorem Garner's Algorithm Scientific Computing CMake Testing