~/portfolio $ whoami

Chandra
Pratap
Singh Chauhan

SDE-2 @ URBANPIPERBENGALURU, INTECH × DESIGN × PRODUCT

I'm a builder — I own tech × design × product end to end. My specialty: backend systems and applied AI — event-driven pipelines, ClickHouse analytics at scale, and LLM agents (RAG, MCP, Claude Agent SDK) that actually ship.

pratap@prod: ~
$ curl -sL singhpratap.dev/api/status | jq '{role, focus, identity, owns}'
{
"role": "SDE-2 @ UrbanPiper",
"focus": ["backend systems", "applied AI / agents"],
"identity": "builder", "owns": "tech × design × product"
}
(that endpoint is real — full resume as JSON)
type `help` to explore ↴
TECH × DESIGN × PRODUCT ✦ BACKEND ENGINEERING ✦ APPLIED AI ✦ CLICKHOUSE ✦ KAFKA ✦ RAG + PGVECTOR ✦ MCP ✦ CLAUDE AGENT SDK ✦ FASTAPI ✦ KUBERNETES ✦ EVENT-DRIVEN PIPELINES ✦

01 / ABOUT

Who I Am

Builder at UrbanPiper (Bengaluru) who owns tech × design × product end to end — from the schema to the pixels to the "why are we building this." Specializes in backend systems and applied AI: event-driven pipelines, ClickHouse analytics, distributed-system consistency, and LLM/RAG in production. Solo-delivered company-wide platforms (engineering on-call intelligence, real-time store-availability analytics) running at scale. Core contributor to Periscope (downtime visibility) and Atlas (merchant dashboard).

// Toolbox

Languages
PythonJavaScriptSQLShell
Databases
ClickHousePostgreSQL/pgvectorMongoDBElasticsearchRedisMySQL
Frameworks
FastAPIReactHasura (GraphQL)DjangoPlaywright
Platforms
KubernetesDockerKafkaGrafanaKibana/ELKSentryPostHog
AI / LLM
RAG (pgvector, embeddings, LLM reranking)MCPClaude Agent SDKPrompt engineeringEvaluation

02 / EXPERIENCE

Where I've Shipped

Software Engineer II

UrbanPiper · Bengaluru, India

Oct 2025 — Present
  • Designed and built, solo, a company-wide AI incident-response system: structured Slack intake with automatic routing, GitHub integration tying each incident to its fix, and a RAG pipeline (pgvector embeddings, LLM reranking, Kibana MCP log retrieval) that surfaces similar past incidents to cut time-to-resolution.
  • Built a coding agent on the Claude Agent SDK that triages critical Sentry alerts, localizes the fault, and opens a reviewable PR with a candidate fix — human-in-the-loop, nothing deploys without review.
  • Built Janus — a dispute-recovery automation service for delivery-aggregator merchant portals (DoorDash, Uber Eats): browser + API automation (Playwright) that files order disputes end to end, guided by a calibrated win-probability model that prioritizes filings under monthly quota limits.
  • Built store uptime/downtime analytics from scratch on ClickHouse (AggregatingMergeTree materialized views) plus the real-time availability layer over a Menu Analyzer processing ~1.5M items / 30 min.
  • Automated store remediation with an "Always-On" recovery layer (anti-flap cooldowns, audited state machine) — 523 successful auto-recoveries and ~$10,200 in revenue saved for merchants to date.
  • Built cross-service sync over Kafka event streaming for real-time multi-service consistency; debugged stuck consumers and ETL pods on Kubernetes.
PythonFastAPIClickHouseKafkapgvector RAGMCPClaude Agent SDKPlaywrightKubernetes

Software Engineer I

UrbanPiper · Bengaluru, India

Jul 2024 — Oct 2025
  • Built a merchant onboarding service that cut average go-live time from ~17 days to ~4 and onboarded ~260 clients — Hasura auto-exposing a GraphQL layer over Postgres, eliminating hand-written CRUD APIs.
  • Architected the alerting system from scratch (event gathering, roll-up, firing) for real-time, de-duplicated merchant notifications of store/item issues.
  • Led the Python 3.6 → 3.9 upgrade of a large legacy service, resolving dependency and compatibility breakage across the codebase.
  • Triaged dozens of SEV-1/SEV-2 production incidents across reporting, analytics, and reconciliation pipelines for enterprise brands in India, the UK, and MENA — including diagnosing silent ingestion data loss and running safe backfills.
  • Delivered Atlas (React) features end to end — a global command palette with fuzzy search (~1,150 opens, ~500 in-product navigations) — plus menu-platform APIs (access control, MENA Menu V2, Lightspeed POS OAuth).
PythonHasuraPostgreSQLReactKafkaKubernetesELK

03 / PROJECTS

Side Quests

Cognify

Cognify

AI-powered assessment platform that turns uploaded documents (SOPs, guidelines) into adaptive, conversational tests — with real-time LLM evaluation of answers.

RAGNext.jsLLM Evaluation
Visit ↗
Powframe

Powframe

AI comic-generation platform transforming stories into comic panels — multiple layouts, style customization, and context-aware narrative flow.

AIImage GenerationNext.js
Visit ↗
Marudhar Resorts

Marudhar Resorts

Website + custom Property Management System (PMS) for a working resort — bookings, rooms, operations — in active production use.

Full-stackPMSProduction
Visit ↗

03.5 / METRICS

Prod Numbers

// live-ish gauges from 2 years in production

ITEMS ANALYZED

0.0 M / 30 min

menu analyzer throughput

MERCHANTS ONBOARDED

~0

go-live: 17 days → 4

STORE RECOVERIES

0

always-on automation

SAVED FOR MERCHANTS

$0

always-on auto-recovery

ON-CALL ISSUES RESOLVED

0 +

production issues · service requests

AI SYSTEMS IN PROD

0

on-call RAG · coding agent · janus

04 / ACHIEVEMENTS

Milestones

🏆 WINNER

Amazon Gen AI Hackathon

Engineered an MCP-backed command palette (PIP) with global fuzzy search, keyboard navigation, and analytics tracking.

🤖 IN PRODUCTION

AI systems that ship

On-call incident intelligence (RAG + MCP) and a human-in-the-loop coding agent — running in production, not demos.

🎓 B.E. CSE

VTU, Computer Science

Bachelor of Engineering in Computer Science, 2020 – 2024.

05 / CONTACT

Let's Talk

Something worth building — backend, applied AI, or the whole tech × design × product loop? A question, an idea, or just want to say hi — my inbox is open.

© 2026 Chandra Pratap Singh Chauhan · built with Next.js · no purple gradients were harmed

NORMAL~/pratap/portfolioBLR IST