KV
varsh@portfolio:~$ whoami
Krishna Varshney
varsh@portfolio:~$ cat role.txt
>
varsh@portfolio:~$ ./current-focus
1M+
Daily Records Ingested
S3, DuckDB & Kafka pipeline
95.8%
Execution Time Reduction
Cut data load from 2h to <5 min
<100ms
Production API Latency
Optimized complex SQL & indexing
$10K+
US Client Projects
Freelance systems & APIs delivered
IncidentPilot — AI-Powered Production Incident Remediation
Autonomous LangGraph Agent
Autonomous multi-agent system that listens for production alerts, performs intelligent AST code search, isolates root causes, synthesizes targeted bug fixes, and submits verified PRs with zero human intervention required until final review.
→
Knowledge-First Code Retrieval: Combines Tree-sitter AST syntax parsing, SentenceTransformers semantic embeddings, call-graph dependencies, and pgvector similarity search to locate exact root-cause files across large codebases.
→
Sandboxed Verification Pipeline: Dispatches generated patches to isolated GitHub Actions runners, guaranteeing that fixes pass all unit & regression test suites before opening pull requests.
→
Security & Guardrails: Enforces HMAC webhook authentication, strict evidence validation, and restricted file-system boundaries for safe autonomous repository interaction.
SalesCode.AI
|
Technical Project Associate (Software Engineer)
Jul 2025 – Present • Coca-Cola Argentina account (40,000+ users) — Best Project of the Year
→Redesigned a 600K-row daily pipeline to delta processing, cutting ingestion time from ~2 hours to under 5 minutes.
→Built a scheduled pipeline processing ~1M records/day across 46 entities, integrating client APIs and S3, transforming via DuckDB, and publishing downstream through Kafka.
→Cut production API response times from ~5s to under 100ms via query profiling, indexing, and SQL optimization.
→Built an incremental Java job matching 7.2M daily records against order/invoice data with bounded memory via batched queries and streamed S3 files.
→Re-architected 20+ Java transformers into a distributed Flink pipeline with entity-level publishing and failed-record isolation for faster recovery.
→Migrated sales-compliance processing from S3 to Amazon Redshift with idempotent upserts, enabling late-arriving orders to update existing records without duplication.
Techentia
|
Freelance Backend Developer
Jun 2024 – Nov 2024 • Remote, US Clients
→Delivered backend engagements worth over $10,000 total for U.S.-based clients, building Node.js/Express APIs for Web3 payment and marketplace platforms.
→Implemented WebSocket-based real-time event feeds for transaction status updates across client platforms.
{
"Languages": [
"Java",
"Python",
"SQL",
"JavaScript",
"TypeScript"
],
"Backend & APIs": [
"Spring Boot",
"FastAPI",
"Node.js",
"Express.js",
"REST APIs",
"WebSockets",
"jOOQ"
],
"AI / Agent Systems": [
"LangGraph",
"LangChain",
"RAG",
"Vector Embeddings",
"Semantic Search",
"Tree-sitter",
"SentenceTransformers",
"OpenAI/Groq API"
],
"Databases": [
"PostgreSQL",
"MySQL",
"MongoDB",
"SQLite",
"DuckDB",
"Amazon Redshift",
"pgvector"
],
"Data Engineering & Messaging": [
"Apache Kafka",
"Apache Flink",
"Event-Driven Architecture",
"Stream & Batch Processing"
],
"System Design": [
"Distributed Systems",
"Idempotent Processing",
"Transactional Outbox",
"Retry & Dead-Letter Handling",
"Fault Tolerance"
],
"Cloud & DevOps": [
"AWS (S3, EC2)",
"Docker",
"Jenkins",
"CI/CD",
"GitHub Actions",
"Git"
]
}
# 01. Engineering Background
Software engineer with 1+ years of production experience specializing in fault-tolerant backend architectures, high-throughput data pipelines, and production-grade LLM agent systems.
Currently at SalesCode.AI on the Coca-Cola Argentina account (40,000+ users), designing distributed master-data ingestion systems processing 1M+ daily records across 46 enterprise entities.
# 02. Core Architectural Invariants
Principles Over Hacks
▹
Zero Dual-Write Races: Enforce the Transactional Outbox Pattern to guarantee event emission consistency between relational stores and Apache Kafka.
▹
Immutable Ledger Balances: Insert-only double-entry bookkeeping ensuring every debit strictly balances a credit with mathematical certainty.
▹
Bounded-Memory Streaming: Process gigabyte-scale datasets using S3 delta diffing and Apache Flink stream pipelines to prevent JVM heap exhaustion.
▹
Sandboxed Agent Guardrails: Autonomous code-remediation agents must execute through isolated CI test gates before human pull request reviews.
# 03. Key Production Milestones
→
95.8% Ingestion Latency Drop: Slashed 600K-row daily master sync from ~2 hours to <5 minutes using S3 delta extraction.
→
50x API Performance Optimization: Reduced mission-critical query latency from ~5s to <100ms via SQL query profiling and compound indexing.
→
7.2M Daily Record Matching: Built streaming Java workers joining client recommendations with bounded memory consumption.
→
Autonomous Incident Remediation: Authored IncidentPilot with LangGraph, Tree-sitter AST parsing, pgvector semantic search, and sandboxed GitHub Actions PR creation.
→
$10,000+ US Client Engagements: Delivered production-ready Node.js, WebSockets, and Web3 smart contract backends for global clients.
# 04. Education & Academics
🎓
ABES Engineering College, Ghaziabad — B.Tech in Computer Science
Nov 2022 – Jun 2026 • CGPA: 8.71 / 10 • Focus on Distributed Systems, DBMS, OS & Algorithms
Nov 2022 – Jun 2026 • CGPA: 8.71 / 10 • Focus on Distributed Systems, DBMS, OS & Algorithms
🏫
DAV Fertilizer Public School — Class XII (CBSE)
Graduated 2021 • Score: 97.2%
Graduated 2021 • Score: 97.2%
# 05. System Status & Availability
Current Role
Software Engineer @ SalesCode.AI
Current Account
Coca-Cola Argentina (Best Project of the Year)
Location
Gurugram, India (Open to Relocation)
Work Preferences
Remote (Worldwide) / Hybrid / On-Site
Target Domains
AI Agents • Distributed Systems • Backend
Contact
varshney.kv1@gmail.com
$ ./hire-me
Let's Build Resilient Systems Together
Open to Software Engineer roles — AI agents, distributed systems, and backend. Relocatable.