Software Engineer · Backend (Java / Spring Boot / Python) · Hands-on AI & LLM integration · Ex-Visa · Bengaluru

I build backend systems for production and make LLMs actually useful inside them: event-driven services, RAG pipelines, realtime voice agents, and guardrails that keep model input and output safe and well-formed.

🔍 Open to full-time and contract SWE roles (backend, AI/LLM, platform). Let's talk on LinkedIn →

Software Engineer · Visa · Bengaluru · Jun 2025 – Aug 2026
Risk & settlement platform

• Built backend services for an AI-assisted conversational feature in Visa's risk exposure tool, so CSR and business teams could query settlement exposure and client limits in natural language over secure REST integrations with production systems
• Shipped an event-driven alert processing service (Java, Spring Boot, IBM MQ) that decoupled alerting from the core settlement platform, cutting alert latency by ~4 minutes
• Engineered a RAG pipeline for duplicate-idea detection that cut LLM context from ~120K to 3–5K tokens, lowering inference cost
• Led migration of a legacy SharePoint platform to Microsoft Dataverse: redesigned the data model and moved 10,000+ production records

Open Source Contributor · Public Invention – Krake firmware · Nov 2025 – present

• 20+ merged PRs to an actively maintained embedded firmware project (C++, ESP32, MQTT): features, debugging and design discussions with maintainers

A browser-based voice agent that holds natural conversations in English, Hindi and Hinglish, answers from a business knowledge base, checks availability and books appointments through real tool calls, then writes a validated, structured post-call summary.

• Streaming STT → LLM → TTS pipeline on LiveKit Agents + Sarvam AI, with barge-in (interruption) support
• Tool calling backed by SQLite; post-call extraction validated with Pydantic
• Provider-agnostic LLM layer and per-stage latency tracking (STT, LLM time-to-first-token, TTS)
• FastAPI backend, Next.js frontend, pytest suite

Python LiveKit FastAPI Next.js SQLAlchemy OpenAI

Middleware that sits between client apps and an LLM provider.

• Inbound: scrubs PII (emails, card numbers) and blocks common prompt-injection / jailbreak patterns
• Outbound: enforces strict Pydantic v2 schemas on model JSON; malformed output returns a structured 502 instead of reaching the client
• Async and non-blocking, with latency reported in X-Response-Time-Ms headers

Python FastAPI Starlette Pydantic

A CLI that reads git history, PRs and migration logs and writes summaries for stakeholders, so PMs and business teams understand what changed and why it matters.

Node.js TypeScript

Languages

Backend & Distributed Systems

AI / LLM

Cloud & DevOps

• Taking VoiceAgent from V0 toward production: telephony (SIP), a RAG knowledge base, multi-tenant support
• Shipping Translayer's core git-commit parser

B.E. Computer Science & Engineering, RNS Institute of Technology, Bengaluru (2021–2025) · CGPA 8.9/10