Instahyre SDE III / Platform Engineer
I'm Abhinav Yadav. I build backend platforms that keep product teams moving.
I work across deploy safety, search migrations, ranking latency, Redis caching, Celery workflows, MySQL scale, and AWS operations. I focus on production paths where correctness, latency, and rollback design matter.
Platform architecture and search systems
SDE III at Instahyre with 6.5+ years across backend, platform, and search. Owns the P0/P1 review gate across 4 teams and 15+ engineers, and builds internal platforms that turn risky production changes into repeatable workflows.
Python
Go
JavaScript
Django
FastAPI
Node.js
AngularJS
Next.js
HTMX
MySQL
PostgreSQL
Redis
Elasticsearch
OpenSearch
Painless
Celery
RabbitMQ
Docker
Nginx
AWS
Datadog
CloudWatch
Sentry
Production systems: Redis, Celery, MySQL
Forge
Internal deployment platform replacing manual SSH deploys with guarded auto-merge, one-command deploy, and rollback. Deploy time fell from 30 minutes to 4 minutes, rollback MTTR from 20+ minutes to under 90 seconds.
ESMultiWrite
Search index migration system for 12+ production indices and 241M+ documents. It made alias swaps, parallel rebuilds, cross-version readers, and fallback paths explicit for the ES 7 to OpenSearch 3 cutover.
Instamatch OpenSearch rewrite
Replaced a proposed caching layer for the busiest candidate ranking API with one OpenSearch query and Painless scoring. p95 dropped from 4.8s to 85ms and app-server load fell about 60%.
BIGINT migration
Led the zero-downtime BIGINT migration of a critical 2B+ row opportunity table using dual-write, online backfill, and atomic cutover while catching bulk-insert deadlocks before referential integrity broke.
Python 2 to Python 3
Sequenced the migration of a 150K+ line legacy codebase over 6 months without blocking feature work and cut over with zero production regressions.
Async and cache hardening
Designed Redis cache-aside patterns, per-key TTLs, event-driven invalidation, Celery queue isolation, retry-storm fixes, and MySQL lock-contention remediation across 100M+ row tables.
Current projects
01 / spaced repetition
LeetDrill
LeetCode spaced-repetition system with SM-2 scheduling, recall scoring, problem history, Chromium extension submission capture, Go backend, PostgreSQL storage, and HTMX review UI.
Go
PostgreSQL
HTMX
02 / search mechanics
Ichnos
Crawler and search engine experiment covering URL scheduling, crawl politeness, HTML cleanup, on-disk inverted index construction, and tunable BM25 ranking.
Crawler
BM25
Indexing
03 / workflow engine
Kleos
Cold-outreach workflow engine for sequences, throttling, suppression lists, reply detection, async worker execution, retries, and per-recipient state isolation.
Async
Queues
Workflow
Project archive
Generates quarterly developer evaluations from Phabricator data.
Static package index for Python wheels on older macOS development machines.
Homebrew tap for bootstrapping and maintaining legacy environments.
Shell-based local setup and development environment configuration.
PHP and MySQL login/register pages, an older public repo with auth basics.
JavaScript object-detection experiment.
React and Express app that converts images to text using Tesseract.
User authentication single-page app using the MERN stack.
Angular drawing app for freehand canvas input.
Flask app that classifies digits drawn on a web canvas using a Keras model.
Small analog clock interface built with React.
Experience
Software Development Engineer III, Platform Architecture
- Designed and shipped Forge, now the only backend deploy path for 15+ engineers and 200+ weekly deployments.
- Replaced the Instamatch caching plan with one OpenSearch query plus Painless scoring, cutting p95 from 4.8s to 85ms.
- Led the zero-downtime BIGINT migration of a 2B+ row opportunity table past its 32-bit primary-key ceiling.
- Built ESMultiWrite for repeatable search index migrations across 241M+ documents and 12+ production indices.
Senior Software Engineer, Scale, Reliability and Modernization
- Decoupled the production runtime into web serving, dedicated Celery worker pools, and outbound processing.
- Designed Redis cache-aside primitives with per-key TTLs, event-driven invalidation, and pooled connections.
- Resolved MySQL replica lag, hot-table deadlocks, CPU saturation, lock contention, and Celery retry storms.
- Led the Python 2 to Python 3 migration of a 150K+ line legacy codebase with zero production regressions.
- Hardened production access with least privilege SSH, Unix isolation, private routing, and AWS Secrets Manager.
Software Engineer, Product Systems and Security
- Shipped Team Management with RBAC, audit trails, impersonation safeguards, and adoption by 1,000+ employer companies.
- Closed authorization bypasses across backend APIs and frontend flows exposing premium features.
- Eliminated race conditions, non-atomic transitions, and full-table scans with Django, Celery, and MySQL fixes.
- Cut incident detection from about 18 minutes to under 9 minutes through Datadog, Sentry, and structured alerts.
Software Engineer Intern
- Built a handwritten digit recognition web app with Flask, Keras/TensorFlow, MNIST, and a JavaScript canvas frontend.
Education
B.Tech, Computer Science
Computer science foundation before moving into backend product engineering, platform architecture, production systems, and infrastructure-adjacent reliability work.