Fig. 1 — a request, on its way down

Aman Tyagi

Senior Backend Engineer

I build the systems you never think about — until they stop working.

Senior Backend Engineer @Delightree
India · IST (UTC+5:30)
amantyagi2k@gmail.com
3+ years shipping production systems
Open to backend & platform roles
he/him

3+ yrs · India · open to offers

Overview

600k+
records migrated
90×
latency reduction
350+
daily automated runs
17
APIs, one domain

Distributed Systems
queues and workers that keep the slow work off the request path
Performance Engineering
took the busiest screen from 4.5 seconds to 50 milliseconds
Data Modeling
reshaped the data so reads cost half as much
AI & RAG
search that answers questions from a company's own documents
API Design
APIs other teams and outside partners build on

Experience

Delightree

Remote

Senior Software Engineer

Full-time06.2025 — present1y 2m

TypeScript
GraphQL
Weaviate
AWS Bedrock
BullMQ
BigQuery
MongoDB
Zod

Multi-tenant RAG engine — Ask a question in plain English, get an answer pulled from the company's own manuals.
no franchise can ever see another's documents
Integration platform — One shared way to connect outside tools — HubSpot, QuickBooks, Zapier and two more.
adding a sixth provider takes an afternoon
Async analytics pipeline — Moved the heavy number-crunching to the background, on a schedule.
reports never slow down the app
Support ticketing domain — Designed and built the whole support-ticket system, start to finish.
a whole support product, built from nothing
Data-model migration — Restructured how the data was stored so the app could find it faster.
halved the database work per read
Partner ingestion API — A documented way for outside companies to send us their data safely.
partners integrate without us writing code
Security hardening — Found passwords leaking into our logs, and stopped it at the source.
closed a credential leak before anyone found it

Playo

Bengaluru, India (On-site)

Software Engineer

Full-time09.2024 — 06.202510m

Node.js
Redis
MongoDB Atlas
TypeScript
AWS

Dense-ranking leaderboard — Precomputed the rankings and kept them in fast memory, instead of rebuilding them every visit.
the slowest screen became the fastest
Geospatial search — Made 'find places near me' actually fast.
search stopped feeling broken
Chat migration — Moved every user's chat history to a new provider without taking the app down.
600k users moved, nobody noticed

Teal India

India (On-site)

Software Engineer

Full-time02.2023 — 08.20241y 7m

Node.js
Socket.io
MongoDB
Express
Docker

Real-time analytics — Live dashboards that kept updating correctly even as we added more servers.
dashboards stayed live as servers scaled out
Legal document-extraction framework — Software that reads legal documents and writes the reports itself.
350+ reports a day, no one touching them

2022 · Where it started: founding engineer at Duckcart, DevOps intern at Recruit CRM.

Incidents

INCIDENT-001

high

600,000 users, two chat providers, one cutover

Full messaging history migration, CometChat → SendBird, for every user on the platform.

01Dual-write to both providers so no message could be lost mid-flight
02Backfill history in batches, checksum every batch against the source
03Cut reads over per-cohort, smallest cohort first
04Verify 100% integrity, then decommission the old provider

Under 50 minutes · 100% integrity · zero downtime.

Dual-write and verify before you cut over. The rollback path is the feature; the migration is just the part that happens if the rollback is never needed.

INCIDENT-002

high

A leaderboard that took 4.5 seconds

The most-viewed screen in the product blocked for 4.5s on every load. Dense ranking over a growing player set, recomputed per request.

01Profile first: the cost was recomputing dense ranks, not fetching them
02Introduce tiered Redis caching keyed by leaderboard segment
03Precompute ranks on write, serve reads from cache
04Add explicit invalidation rather than short TTLs, to keep reads correct

4.5s → 50ms. A 90× reduction on the hot path.

Measure before optimizing. The obvious suspect was the database; the actual cost was work being redone on every single request.

INCIDENT-003

moderate

Credentials in the log pipeline

Application logs being shipped to an external aggregator included credential material that should never have left the process.

01Identify every field carrying sensitive values through the shipper
02Redact at the source, before serialization, not at the sink
03Rotate the exposed credentials
04Re-audit the shipped payload to confirm the leak was closed

Zero secrets in shipped logs, verified after the fix.

Redact where the data is produced, not where it is stored. A sink-side filter is one config change away from leaking again — and audit what you ship, because nobody reports a leak they cannot see.

Stack

core

JavaScript / TypeScript
Node.js / Express
GraphQL (Apollo)
Distributed Patterns

data

MongoDB
Aggregation / Indexing
BigQuery
Weaviate

async

Redis
BullMQ
Cron
Webhooks

platform

AWS
Docker

RAG
Vector Search
AWS Bedrock

Contact

Open to backend and platform engagements. Based in India, working across timezones.

Email Aman Resume

Built with Next.js.

Overview

600k+

records migrated

90×

latency reduction

350+

daily automated runs

APIs, one domain

Distributed Systems

queues and workers that keep the slow work off the request path

Performance Engineering

took the busiest screen from 4.5 seconds to 50 milliseconds

Data Modeling

reshaped the data so reads cost half as much

AI & RAG

search that answers questions from a company's own documents

API Design

APIs other teams and outside partners build on

Experience

Delightree

Remote

Senior Software Engineer

Full-time06.2025 — present1y 2m

TypeScript
GraphQL
Weaviate
AWS Bedrock
BullMQ
BigQuery
MongoDB
Zod

Multi-tenant RAG engine — Ask a question in plain English, get an answer pulled from the company's own manuals.
no franchise can ever see another's documents
Integration platform — One shared way to connect outside tools — HubSpot, QuickBooks, Zapier and two more.
adding a sixth provider takes an afternoon
Async analytics pipeline — Moved the heavy number-crunching to the background, on a schedule.
reports never slow down the app
Support ticketing domain — Designed and built the whole support-ticket system, start to finish.
a whole support product, built from nothing
Data-model migration — Restructured how the data was stored so the app could find it faster.
halved the database work per read
Partner ingestion API — A documented way for outside companies to send us their data safely.
partners integrate without us writing code
Security hardening — Found passwords leaking into our logs, and stopped it at the source.
closed a credential leak before anyone found it

Playo

Bengaluru, India (On-site)

Software Engineer

Full-time09.2024 — 06.202510m

Node.js
Redis
MongoDB Atlas
TypeScript
AWS

Dense-ranking leaderboard — Precomputed the rankings and kept them in fast memory, instead of rebuilding them every visit.
the slowest screen became the fastest
Geospatial search — Made 'find places near me' actually fast.
search stopped feeling broken
Chat migration — Moved every user's chat history to a new provider without taking the app down.
600k users moved, nobody noticed

Teal India

India (On-site)

Software Engineer

Full-time02.2023 — 08.20241y 7m

Node.js
Socket.io
MongoDB
Express
Docker

Real-time analytics — Live dashboards that kept updating correctly even as we added more servers.
dashboards stayed live as servers scaled out
Legal document-extraction framework — Software that reads legal documents and writes the reports itself.
350+ reports a day, no one touching them

2022 · Where it started: founding engineer at Duckcart, DevOps intern at Recruit CRM.

Incidents

INCIDENT-001

high

600,000 users, two chat providers, one cutover

Full messaging history migration, CometChat → SendBird, for every user on the platform.

01Dual-write to both providers so no message could be lost mid-flight
02Backfill history in batches, checksum every batch against the source
03Cut reads over per-cohort, smallest cohort first
04Verify 100% integrity, then decommission the old provider

Under 50 minutes · 100% integrity · zero downtime.

Dual-write and verify before you cut over. The rollback path is the feature; the migration is just the part that happens if the rollback is never needed.

INCIDENT-002

high

A leaderboard that took 4.5 seconds

The most-viewed screen in the product blocked for 4.5s on every load. Dense ranking over a growing player set, recomputed per request.

01Profile first: the cost was recomputing dense ranks, not fetching them
02Introduce tiered Redis caching keyed by leaderboard segment
03Precompute ranks on write, serve reads from cache
04Add explicit invalidation rather than short TTLs, to keep reads correct

4.5s → 50ms. A 90× reduction on the hot path.

Measure before optimizing. The obvious suspect was the database; the actual cost was work being redone on every single request.

INCIDENT-003

moderate

Credentials in the log pipeline

Application logs being shipped to an external aggregator included credential material that should never have left the process.

01Identify every field carrying sensitive values through the shipper
02Redact at the source, before serialization, not at the sink
03Rotate the exposed credentials
04Re-audit the shipped payload to confirm the leak was closed

Zero secrets in shipped logs, verified after the fix.

Redact where the data is produced, not where it is stored. A sink-side filter is one config change away from leaking again — and audit what you ship, because nobody reports a leak they cannot see.