About

I am a forward-deployed AI engineer in Bengaluru. Most of my work sits between the model and the product — and often between the product and the client.

Forward-deployed means staying close to the workflow: translating a messy client process into prompts, tools, integrations and a release process that survives real call traffic. The interesting part is rarely the model call. It is the retrieval, orchestration, latency budget and failure path around it.

At NuPlay AI I have launched production voice agents for solar, real estate, fitness and restaurant-ordering campaigns, from campaign setup and CRM integration through UAT and post-launch support. Before that I built backend APIs and schemas at Uflex.

Independently, I built GeoLLM, a planner that turns plain-language questions into satellite analysis, tldrepo, a local-first CLI that maps a codebase with a small model, and Prompt2Shell, a fine-tuned model that writes shell commands. I am drawn to narrow interfaces that keep their evidence visible.

Experience

Jan 2026 — PresentBengaluru, India

Forward Deployed Engineer (AI Voice Agents)

NuPlay AI

Forward-deployed on four production LLM voice-agent launches, owning campaign setup, integrations, release verification and the reliability work after go-live.

Ran launches for solar, real estate, fitness and restaurant ordering — campaigns of 2K–40K leads — with CRM and webhook integration, UAT, and a backup, draft-test, publish and live-verify release process.
Restaurant ordering: built a serverless AWS Lambda API that updates the menu section of the live prompt on demand, replacing a per-order menu fetch; LLM dish understanding replaced similarity matching and caught native Marathi and South Indian dishes.
Solar: launched the campaign and a second region on a shared agent with Conductor-based workflow automation — LLM address-cleaning and a city-routed post-call branch — producing 540 visits from 2K leads and 3K call attempts (27%).
Real estate: 20K leads and 65K call attempts; shipped client-requested prompt releases for pricing, call flow and language guards, verified by diff and re-fetch, plus an available-now callback that feeds an email-alert workflow.
Fitness: pushed 40K leads with 56% answered; promoted the tested STT/TTS/LLM configuration to production with a checklist and post-change verification, and tuned voice and call-screener handling.
Reporting and observability: Metabase and SQL dashboards for call, lead and visit-confirmation metrics, plus Langfuse trace analysis for root-cause debugging of live agents.

Python · FastAPI · AWS Lambda · Conductor · Soniox · Cartesia · ElevenLabs · Langfuse · Metabase

Jul 2025 — Nov 2025Noida, India (Remote)

Software Development Engineer Intern

Uflex

Backend work on an internal Issue Tracker platform used by cross-functional teams.

Designed and built backend REST APIs and database schemas for issues, comments, notifications and lifecycle transitions.
Debugged backend services and improved performance through testing and code refactoring.

Python · REST APIs · SQL

Education

2022 — 2026Mohali, India

B.E. Computer Science (AI & ML)

Chandigarh University · CGPA 8.2

Graduated May 2026. Certifications: Oracle Cloud Infrastructure Generative AI (Jun 2024); Microsoft Introduction to Generative AI (Mar 2025).

Toolkit

Languages & backend

Python, SQL, JavaScript, TypeScript, FastAPI, Flask, REST API design, SSE streaming, React, Next.js

Generative AI

LLMs, prompt engineering, RAG, FAISS, SentenceTransformers, QLoRA fine-tuning, tool calling, LangChain, ROUGE/BLEU evaluation, guardrails, NLP

Voice & LLMOps

Voice agents, STT/TTS (Soniox, Cartesia, ElevenLabs), Langfuse tracing, observability, post-call analytics, root-cause analysis

Data & cloud