Headline

AI Engineer at Yarnit

About

I build AI systems that actually ship to production — not demos, not POCs, not "coming soon." In 2 years at KPIT Technologies, I've architected and delivered Generative AI solutions for three of the world's most iconic automakers: Ford, Jaguar Land Rover, and Honda. Every system I've built runs in production, serves real users, and solves hard engineering problems. Here's what that looks like in practice: 🔹 I'm building AI Intel Query, an Agent-to-Agent (A2A) NLP platform that translates natural language into precise SQL queries across multi-gigabyte network datasets. Think: "Show me all failed routing events on Bus X last Tuesday" → instant, accurate results. No SQL knowledge required. 🔹 I engineered a data pipeline processing 1M+ network packets with anomaly detection that flags intrusions and replay attacks in real time. I also integrated an LLM-powered RAG system enabling plain-language queries over complex automotive network behavior. 🔹 I architected a Multimodal RAG platform that ingests text, charts, tables, and equations from complex engineering documents, and a Video Intelligence system that cut manual documentation time by 90% and LLM token costs by 40%. My technical skills: → Generative AI: LLMs, Multimodal RAG, Agentic AI, A2A, Knowledge Graphs, Prompt Engineering → Backend: Python, FastAPI, AsyncIO, WebSockets, PostgreSQL, Redis → Frontend: React 19, TypeScript, Zustand, ECharts → Data: Polars, DuckDB, Pandas, Apache Parquet, scikit-learn What drives me isn't just the tech — it's watching an engineer at Ford type a question in plain English and get an answer in seconds that used to take hours of manual trace analysis. That's the kind of impact I want to keep building. I'm always interested in connecting with people working on: • Production AI/ML systems (especially RAG and agents) • High-throughput data engineering challenges Reach out if you're building something ambitious with AI — or if you just want to geek out about Gen AI. I'd love to chat.

Location

Pune Division

AI stack

antigravity

claude code

claude

codex

cursor

deepseek

Work experience

AI Engineer at Yarnit, Sep 2026 - Present · 3 mos, Bengaluru, Karnataka, India

Generative AI Engineer at KPIT, Nov 2025 - Aug 2026 · 10 mos, Pune District, Maharashtra, India, Built AI Intel Query an A2A natural language platform that lets engineers query multi-gigabyte network datasets in plain English instead of writing SQL. Powered by a multiRAG architecture with hybrid vector-graph retrieval, enabling instant answers from datasets that previously required hours of manual trace analysis. Engineered a high-throughput data pipeline processing massive PCAP network captures (1M+ packets per trace) using Polars and DuckDB's zero-copy analytical engine enabling instant data retrieval from multi-gigabyte datasets without exhausting server memory. Developed a secure, multi-tenant analytics dashboard (React 19, TypeScript, Zustand) with role-based access control enabling stakeholders to see high-level routing insights while testers dive into granular packet-level analysis through interactive Sankey flows and heatmaps.

Generative AI Engineer at KPIT, Nov 2024 - Oct 2025 · 1 yr, Pune District, Maharashtra, India, Built Packet Predator, an automotive network intelligence platform for Honda that processes and analyzes CAN/CAN-FD/Ethernet traffic at scale, with real-time anomaly detection and natural language querying capabilities. Built a real-time signal processing pipeline handling 1M+ automotive network packets (CAN/CAN-FD/Ethernet) via async FastAPI with WebSocket streaming, Redis caching, and PostgreSQL delivering sub-second analysis that previously required batch processing. Deployed ML-based anomaly detection to automatically flag network intrusions and replay attacks in automotive traffic then integrated an LLM-powered RAG pipeline so Honda engineers could query network behavior in plain English instead of parsing raw packet logs.

Generative AI Engineer at KPIT, Aug 2024 - Oct 2024 · 3 mos, Pune District, Maharashtra, India, Built a Video Intelligence system for Jaguar Land Rover that combines multimodal LLM analysis with speech recognition to enable semantic video search and automated report generation dramatically reducing manual documentation effort. Built a Hybrid Video RAG Engine combining multimodal LLM frame analysis with Whisper-based speech recognition enabling JLR engineers to semantically search through hours of test footage at sub-second latency instead of manually scrubbing video files Reduced manual documentation time by 90% through automated report generation and cut LLM token costs by 40% via dynamic context assembly and intelligent caching turning a tedious multi-hour documentation process into a one-click operation

Education

B.Tech, Electrical, Electronics and Communications Engineering, Pimpri Chinchwad Education Trust'S. Pimpri Chinchwad College Of Engineering