CV
Lead ML Engineer with four years building distributed, high-throughput production systems at enterprise scale.
Experience
Lead Machine Learning Engineer, AI/ML
Sprinklr, Gurugram. July 2022 to presentArchitecting agentic workflow copilots that automate dashboard analytics through natural language: summarisation, query routing and conversational Q&A.
Unified multi-LLM request router
- Solo-designed and built a microservice that puts Azure, OpenAI, Gemini and in-house models behind one API for HTTP and Kafka workloads, used across Sprinklr’s AI products.
- Drove $100,000+ in annual savings with granular usage telemetry and automated mitigation.
- Built for fault tolerance at 1M+ requests a day with 99.9% uptime.
DeepAgents for an AI content generation copilot
- Led the brand understanding layer: structured brand profiles that ground downstream agent reasoning and content generation.
- Built DeepAgents for insight generation from social signals, engagement metrics and external web context.
- Integrated contextual data sources and retrieval into agent workflows for brand, audience and market context.
Workflow-based agentic dashboard copilot
- Built a DAG-based orchestration layer routing queries across 200+ analytics sources, with Python REPL tool execution and LLM function calling.
- Built an LLM-as-judge framework to score agent quality and track prompt optimisation across model upgrades; it caught a live regression after an upstream OpenAI model update.
- Implemented sliding-window context management with session-level state for coherent multi-turn conversations.
- Owned an end-to-end summarisation system in a five-person team, distilling insights from 400+ dashboard widgets and cutting time-to-insight by 70%+.
- Architected a custom field integration layer for user-defined metrics with non-standard naming.
Platform infrastructure and mentoring
- Contributed to an internal AutoML platform (100+ training requests a day), cutting model development time by 50%+.
- Designed a distributed rate limiting SDK in Python (Redis + Lua) with zero-downtime config refresh via MongoDB.
- Built a secure HTTP + gRPC proxy that centralised MongoDB access org-wide, removing ~1,000× direct connections.
- Saved ~$10,000 a month with Nvidia GPU time slicing; built a GenAI Python SDK that generates unit tests.
- Mentored engineers from a junior engineer and interns early on to two or three concurrent junior engineers.
Machine Learning Intern
Sprinklr, Gurugram. May to July 2021- Developed a central platform to manage Sprinklr’s ML models and datasets.
- Fine-tuned RoBERTa on Hinglish and Hindi datasets for multilingual classification.
Award
Sprinklr Manager’s Spotlight Award (Q3 FY24), for the reliability and robustness of core ML services.
Open source
Mercari AI Shopper
Multi-agent AI shopping assistant- Five sequential agents decompose natural-language queries into Mercari Japan search strategies, scrape dynamic pages with Playwright, and filter candidates.
- RAG over scraped listings grounds recommendations in real data, with ranked results and reasoning.
- Selective tool calling: native function calling only where validation matters, JSON parsing elsewhere.
- Fully async pipeline with batch semaphores; English and Japanese; 90+ unit tests.
Skills
- AI agents and orchestration
- DeepAgents, LangChain, LangGraph, agentic workflows, tool use and function calling, multi-agent systems, prompt engineering and optimisation, multi-LLM routing, RAG, LLM-as-judge evaluation
- Infrastructure and tools
- Azure, GCP, Docker, Kubernetes, Kafka, Redis, Elasticsearch, MongoDB, gRPC, MLOps
- Languages
- Python (primary), Java
Education
Delhi Technological University
August 2018 to May 2022B.Tech in Computer Science. CGPA 9.20/10.