AI/ML Engineer / Building production AI systems
Specializing in production-grade agentic AI, voice pipelines, computer vision, time-series forecasting, and backend systems at scale.
I enjoy building production AI systems rather than research prototypes. My work sits at the intersection of large language models, real-time voice AI, computer vision, and the backend infrastructure that makes these systems reliable in production.
I joined Zypp Electric as an early member of the AI team and have since built AI systems that support operations serving 25,000+ active delivery riders across India.
To me, building an AI system is about much more than training a model. It needs to handle edge cases gracefully, fail safely, and remain reliable in real-world conditions. Whether it's a voice assistant operating in noisy environments or a credit scoring engine that needs to be consistent across different cities, I believe the engineering behind an AI system is just as important as the model itself.
What I Focus On
LLMs & Agentic AI
Tool-calling systems, streaming inference, RAG pipelines
Computer Vision
Damage detection, OCR, and face verification pipelines
Voice AI
Real-time multilingual voice assistants with WebRTC
Time-Series ML
Demand forecasting with foundation models
Backend Systems
FastAPI async services, Redis, and MySQL for high-throughput workloads
Each project is a case study in real-world AI engineering — from problem statement to architecture to production impact.
Multi-VCS AI-Powered Regression Testing & Automated Code Review Platform
Automates multi-VCS pull request reviews, monorepo AST impact mapping, parallel agent execution, sandboxed pytest runs, and generative test creation.
Multi-VCS Support
VCS Integration
Parallel LangGraph
Agent Orchestration
Close to 32% Token Savings
Context Pruner
Isolated Pytest
Execution Sandbox
Real-Time Multilingual Voice AI Assistant
Deflected 40%+ of support calls by automating earnings and wallet query resolutions over voice.
25,000+
Riders Served
Hindi, English + Indic
Languages
< 300ms
RAG Retrieval
5 (B2C, B2B, RTO, Loader, Onboarding)
Roles Supported
Streaming Conversational AI for Riders & Team Leaders
Automated 50%+ of daily worker support tickets through natural language indicating systems.
2.0 Live, 3.0 in Testing
Phase
200 req/min
Request Rate Limit
B2B + B2C (role-aware)
Models Supported
Hindi, English + Indic
Languages
Automated Damage Detection with GPT-4o Vision
Saved 100+ manual operations hours/day by validating damage checks in under 12 seconds.
4 (parallel)
Sides Processed per Ride
100 req/min
API Rate Limit
GPT-4o (temp 0.1)
Vision Model
5-point per image
Quality Checks
Architectural concepts and systems in active design — not yet shipped to production.
Autonomous AI Engineering Organization as a Service (Idea)
Conceptualized modular agent pipeline to orchestrate multi-agent microservices.
DAG-based WorkflowManager
Agent Orchestration
Token budget circuit breaker
LLM Loop Safety
2-Layer Strategy Drift Checks
Drift Verification
n8n Tool Bridge Support
Integrations
Technologies I use in production, organised by domain. Tiers reflect hands-on production use - verified by the projects listed below each domain.
Used in
A summary of my impact at Zypp Electric (India's largest EV-first last-mile delivery platform).
Zypp Electric
Real-Time Voice PipelineScaled and deployed Zypp Saathi voice assistant for 25,000+ active riders using Pipecat, OpenAI Realtime API, and LiveKit WebRTC.
Agentic Framework MigrationArchitected the B2C & B2B conversational core in Zia Chatbot (Phase 3.0) using LangGraph state machines.
High-Performance CRM IngestionDeveloped a robust FastMCP (Model Context Protocol) server for real-time lead captures with duplicate filters.
Deep Vision & OCR SystemsIntegrated Nvidia Nemotron-Parse for EV component image OCR to automate parts validation.
In my full-time engineering role, my focus shifted to system design, throughput scaling, and cost containment. I worked on reducing API latency over WebSockets, managing concurrent streaming token states, and building secure server-to-server connection layers.
Zypp Electric
Automated Damage DetectionMastered vision parsing by building a GPT-4o Vision validation pipeline to analyze 4 vehicle sides in parallel for ride-ends.
Proprietary Scoring SystemsEngineered Hustle Score (Zybil), mapping EV IoT telemetry, wallet records, and merchant stats into a 300-900 scale for gig worker credit.
Demand Forecasting PipelinesImplemented spare parts consumption forecasting across 15+ cities using Amazon Chronos-2 with statistical TSB/SBA fallbacks.
LLM Orchestration & RAGDeveloped initial runs of the Zia Chatbot, utilizing Qdrant Vector DB context injections and low-latency NDJSON streaming.
During this year-long internship, I transitioned from experimental models to shipping production-grade AI pipelines serving thousands of riders daily. I gained hands-on experience in managing LLM pricing boundaries, system latency bottlenecks, asynchronous task workers, and structuring strict validation checks to prevent data leaks. Converted to full-time AI/ML Engineer in July 2026 after shipping 5 production systems across voice AI, computer vision, credit scoring, forecasting, and LLM orchestration during the internship.
I am an AI/ML engineer focused on building production-grade agentic architectures and intelligent systems in production. I specialize in designing multi-agent workflows, real-time voice pipelines, computer vision, and forecasting models. I leverage strong backend engineering and modern ML frameworks to solve real-world automation challenges in production environments.
Location
Gurgaon, India
Education
B.Tech in Computer Science and Engineering
B.Tech in Computer Science and Engineering
Focused on computer science core, software systems, and artificial intelligence. Built the technical foundation for applied AI and backend engineering.
Intermediate Education (Class XII)
Completed Senior Secondary Education with a focus on Physics, Chemistry, Mathematics, and Computer Science.
Secondary Schooling (Class X)
Completed high school with distinction.
I'm open to collaborations and interesting conversations about AI engineering.