~/piyush· 

I build multi-agent AI platforms and the backends that power them.

Backend & AI platform engineer at WizCommerce — agent orchestration, MCP tool platforms, RAG pipelines, and multi-tenant LLM infrastructure for enterprise commerce. Python and Java, from prompt to production.

3+

years building backend & AI systems

200+

AI tools live across enterprise domains

~90%

LLM schema-token overhead eliminated

01 · Focus

AI platform work

The systems I spend my days on — the layer between large language models and enterprise backends.

01

Multi-Agent Orchestration

Domain-specialist subagents on LangGraph with dynamic routing, per-tenant graph caching, and context compression — one agent pipeline serving both conversational and machine-readable workflows.

LangGraphLangChainLiteLLM

02

MCP Tool Platforms

Enterprise backends exposed as typed, tenant-aware AI tools — 200+ callable tools across CRM, commerce, and analytics, with input validation and ~90% schema-token compression.

Model Context ProtocolFastAPIPydantic

03

RAG & NL-to-SQL

Production retrieval pipelines: hybrid BM25 + dense vector search with RRF, Voyage reranking, SQL self-correction, and read-only BigQuery guardrails for analytics over natural language.

QdrantVoyage AIBigQuery

04

Guardrails, HITL & Evals

Human-in-the-loop confirmation for destructive actions, per-tool policy enforcement, PII scrubbing, LangSmith tracing, and domain-specific eval suites for regression-safe agents.

LangSmithLangfuseAI Guardrails

02 · Experience

Where I've worked

Software Development Engineer I (Backend & AI)

WizCommerce (Oritur Technologies Pvt Ltd) · Bengaluru

Sep 2025 — Present

  • Co-architected WizCommerce's shared AI platform — reusable agent orchestration, LLM infrastructure, and enterprise AI services behind WizPilot, WizStudio, Ella, and Workflow Engine.
  • Built WizPilot, the in-app agentic assistant connecting WizOrder with CRM, catalog, cart, analytics, and order management — sales teams run end-to-end B2B workflows in natural language.
  • Designed a database-configured, multi-tenant agent execution layer with per-tenant graph caching, PostgreSQL-backed conversation state, and dynamic MCP/subagent registration.
  • Built a reusable MCP platform exposing 200+ typed, tenant-aware tools; dynamic tool compression cut LLM schema-token overhead by ~90%.
  • Shipped a production RAG NL-to-SQL pipeline (retrieve → generate → validate → secure → execute) with Qdrant retrieval, Voyage reranking, and read-only BigQuery guardrails.
  • Built a tenant-scoped Knowledge Base with hybrid BM25 + dense retrieval (RRF), SSE streaming of tokens, tool calls and reasoning, agent skills, and cross-thread long-term memory.
  • Implemented guardrails and human-in-the-loop controls — confirmation flows for destructive mutations, per-tool policies, and PII scrubbing — for safe, auditable AI actions.
  • Cut critical endpoint latency by 80%, reduced system errors by 32%, and built a parallelized migration pipeline moving 1M+ records at 1,000+ docs/sec.
PythonFastAPILangGraphMCPLiteLLMPostgreSQLQdrantBigQueryElasticsearchRedis

Software Development Engineer I (GenAI)

Fynd (Shopsense Retail Technologies Ltd) · Mumbai

Feb 2025 — Aug 2025

  • Led backend development for the Fynix AI coding assistant powering the IntelliJ and VS Code extensions and the Fynix web platform.
  • Improved codebase indexing speed by 90%+ through optimization of embedding models, vector databases, and keyword/vector search.
  • Designed ETL pipelines integrating GitHub, GitLab, Azure DevOps, Quip, Google Docs, and BigQuery; ingested Stack Overflow and public GitHub datasets to enrich LLM context.
  • Expanded Fynix from a coding assistant into a data-analysis agent using LangChain and LangGraph; boosted accuracy with intent detection and improved RAG workflows.
  • Implemented full-stack observability via Sentry, Prometheus, Grafana, and New Relic; ensured SOC 2 and GDPR compliance.
PythonJavaKotlinLangChainLangGraphVector DBOpenAIClaude

Backend Developer (Founding Team)

Growder Technovations & Ompax Lifestyle · Surat

Mar 2023 — May 2024

  • Engineered high-performance backend architectures across two early-stage B2B platforms using Java, Spring Boot, and distributed microservices.
  • Designed a large-scale Order Management System and production-grade user management with JWT auth and RBAC — throughput up 70–80%.
  • Built configuration-driven integrations with Shiprocket, Shopify, GST/PAN verification, payment gateways, and SMS/email providers.
  • Automated warehousing, logistics, and inventory through event-driven microservices — operational efficiency up 25%, reporting 40% faster.
JavaSpring BootMySQLPostgreSQLElasticsearchAWSRabbitMQ

03 · Skills

Tools of the trade

AI & LLM

Multi-Agent SystemsAgentic AIRAGModel Context Protocol (MCP)Prompt EngineeringTool CallingNL-to-SQLHybrid Search (BM25 + Vector / RRF)RerankingAI Guardrails & HITLLLM EvaluationLangSmithLangfuseVoyage AIOpenAIAnthropic ClaudeGoogle Gemini

Languages & Frameworks

PythonJavaJavaScriptSQLFastAPILangChainLangGraphLiteLLMSpring BootSQLAlchemyAlembicPydanticPytestJUnit

Cloud & DevOps

AWS (S3, EC2, Lambda)GCP (BigQuery, Cloud SQL, GCS)DockerKubernetesCI/CDGrafanaPrometheusSentryNew Relic

Data & Messaging

PostgreSQLMySQLRedisQdrantElasticsearchKafkaRabbitMQKibana

Engineering

Multi-Tenant ArchitectureMicroservicesStreaming APIs & SSEREST APIsAPI GatewayHigh-Level DesignLow-Level DesignData Structures & Algorithms

ML & Data

Supervised & Unsupervised LearningFeature EngineeringModel EvaluationScikit-LearnEDAStatistical AnalysisTableauMatplotlib & Seaborn

04 · Education

Degrees & certifications

2025

Master of Computer Applications (MCA)

Machine Learning and Artificial Intelligence

Amity University, Noida

2020

Bachelor of Business Administration (BBA)

Birla Institute of Technology, Mesra

2025

Certification

Data Science and Machine Learning

Scaler DSML

2023

Certification

Full Stack Web Development

Masai School, Bengaluru

05 · GitHub

Open activity

fetching github stats…

06 · Connect

Let's build something intelligent.

Open to conversations about AI platforms, agent systems, backend architecture — or interesting problems in general.

Say hello →

Location

Bengaluru, Karnataka

© 2025 Piyush Tyagi · built with Next.js