SHUBHAM SINHA
0
Shubham Sinha
Resume Let's talk
Available for Python backend & AI engineer roles

Backend engineer building systems that search, sync & scale.

I'm Shubham Sinha — a Python backend & AI engineer with 5+ years building fast APIs and search infrastructure, real-time data pipelines, and shipping ML into products — from catalog classification to RAG.

View my work Watch intro Download resume
GitHub LinkedIn LeetCode
01

About

90-second intro
Portrait of Shubham Sinha
5+ years
building

I build the unglamorous parts of products that make them actually fast — search that finds the right thing, pipelines that keep data fresh, and services that stay quick under real load.

At Numerator I optimize BI query paths over a billion-row store, run a layered Django REST + Elasticsearch service, and stream MySQL changes into search with a CDC pipeline that replaced overnight reindexes with sub-minute sync. Earlier, at Inexture, I shipped UPI deposit flows for a crypto platform serving 19M+ users and built the observability that cut incident MTTR.

I care about correctness under concurrency, tight types, and systems you can reason about at 2am. When I'm not on infra, I build things like CricLot — a real-time, server-authoritative auction engine.

50% higher analytical throughput
<60s MySQL → search sync
19M+ users served on payments
02

Experience

Jan 2024 — Present Remote, India

Engineer 1 · Numerator

  • Optimized BI query paths on a billion-row MySQL store, cutting heavy analytical reads from minutes to seconds and raising throughput 50%.
  • Engineered a layered Django REST service over Elasticsearch, splitting business APIs, search & ingestion — median API latency down 150ms.
  • Built MySQL→Elasticsearch CDC with pub/sub + Celery, replacing overnight reindexes with sync under 60s.
  • Automated EC2, RDS, S3 & Lambda provisioning with Terraform, cutting env setup from hours to minutes.
  • Fine-tuned Hugging Face models with spaCy preprocessing (US/CA/DE catalogs), improving brand/category attribution 23%.
PythonDjango RESTElasticsearchCeleryTerraformAWS
Dec 2020 — Dec 2023 Ahmedabad, India

Python Developer · Inexture Solutions

  • Delivered UPI deposit workflows for a crypto platform serving 19M+ users, enabling in-app wallet funding without card rails.
  • Launched Grafana/ELK payment dashboards with on-call teams, cutting MTTR 40% via earlier anomaly detection.
  • Engineered RabbitMQ producers/consumers that decoupled synchronous paths across microservices on a national employment platform.
  • Streamlined multi-stage builds & deployments with Jenkins CI/CD, removing most manual release steps.
PythonRabbitMQGrafanaELKJenkins
03

Education

Aug 2017 — May 2021 Nadiad, Gujarat, India

Bachelor of Technology in Information Technology · Dharmsinh Desai University

Information TechnologyB.Tech
04

Selected Work

Featured project

Incident Commander

An SRE multi-agent copilot on LangGraph 1.0. A production alert fans out to specialist agents (logs, metrics, changes, runbooks), a planner synthesises a diagnosis, and the graph interrupts before any destructive action — then resumes from a durable checkpoint after human approval.

  • Parallel investigation subgraph with custom reducers for safe fan-out merge
  • Human-in-the-loop via interrupt-before + update-state resume — no work repeated
  • MemorySaver in dev, Postgres checkpointer in prod; SSE streaming + PagerDuty/Datadog webhooks
LangGraphLangChainFastAPISSEPostgresHITLLangSmith
Read case study View source Architecture wiki
Alert
Agents
HITL
Execute
Featured project

PulseSearch

A real-time, AI-native search system over a live data firehose. Change data capture streams every row change from MySQL through Debezium + Kafka into Elasticsearch, served as hybrid BM25 + vector search and a grounded RAG assistant (Groq). Compose for the full demo; Kubernetes for the app tier; GitHub Actions → GHCR for CI/CD.

  • Effectively-once indexing on at-least-once delivery via version-guarded upserts, replayable backfill & dead-letter isolation
  • Hybrid retrieval: BM25 + kNN fused with Reciprocal Rank Fusion, with per-hit rank provenance
  • K8s Deployments with probes + Ingress; CI with Ruff, Bandit, ES smoke, and image publish to GHCR on main
FastAPIKafkaDebeziumElasticsearchGroqRAGKubernetesGitHub Actions
Read case study View source Architecture wiki
Firehose
CDC
Index
RAG
Featured project

CricLot

A live cricket player auction platform — host IPL-style private auctions with team budgets, real-time bidding, and host/team/display consoles. One server is the source of truth for the timer, current lot, purses and squads, with a WebSocket hub keeping every screen in sync — no refresh hacks.

  • Server-authoritative auction engine + lot state machine (hold timer resets on bid)
  • Concurrency-safe bidding: validates purse, increment & squad cap; persists every bid
  • Bid fan-out to 50 sockets in ~5 ms p50 / 16 ms p95, verified by an E2E suite in CI
  • Server-side host / team / display roles; deployable to Fly / Railway + Vercel
FastAPINext.js 15WebSocketsSQLAlchemyTypeScriptDockerGitHub Actions
Read case study View source Architecture wiki
Host
WS Hub
Engine
Teams
🔎
Case study →

Catalog Search Platform

Product search API on FastAPI + Elasticsearch (BM25 + dense vectors) where keyword-only search kept missing brand & category variants. A Hugging Face classifier on ingest lands new products under the right facets automatically.

+23% brand / category attribution accuracy

FastAPIElasticsearchHugging Face
🔁
Case study →

Near Real-Time CDC Search Sync

Streamed MySQL changes through Debezium + Kafka into Elasticsearch with idempotent consumers, retries, and a replayable backfill — replacing overnight full reindexes with continuous, consistent sync.

~12h → <60s search sync latency

DebeziumKafkaElasticsearch
📈

Stock Analysis Crew AI

A multi-agent research assistant that coordinates specialized agents to gather, analyze, and summarize equities data into a readable investment brief.

PythonCrewAILLMs
🩺

Medical Chatbot

A retrieval-augmented chatbot that answers health questions from a curated knowledge base, grounding responses to reduce hallucination.

PythonRAGNLP
See all repositories on GitHub
05

Skills & Tools

Languages

PythonSQLGo

Backend & Data

DjangoDRFFastAPIMySQLPostgreSQL ElasticsearchCeleryRabbitMQKafkaDebezium

Cloud & Delivery

AWSTerraformDockerKubernetesJenkinsGitHub ActionsGrafanaELK

AI / NLP

LangGraphLangChainHugging FacespaCyTransformersRAG
06 — Contact

Let's build something
fast.

Open to Python backend & AI engineer roles. The quickest way to reach me is email — I usually reply within a day.

Click to copy Or open in Gmail
GitHub LinkedIn LeetCode Resume