Jathin Shettigar

About

Full-stack engineer who built an enterprise search platform (Codlens) delivering natural-language Q&A over Jira/Slack using embeddings, vector DB search, re-ranking (RRF), and LLM responses with source grounding. Also designed and benchmarked a distributed IAM system with Postgres transaction-log replication and Raft-based quorum consistency, reporting ~253 TPS at ~60ms latency in a multi-node setup. Experience spans early-stage startups (Zetic AI, Sagwara Capital) and large-scale orgs (Akamai, Atlassian).

Key Strengths

• Shipped end-to-end enterprise semantic search + Q&A feature (Next.js App Router + FastAPI) integrating Jira/Slack
• Performance optimization via pipeline-level latency measurement (embedding, vector search, LLM stages)
• Vector search tuning (index tuning, limiting search scope) plus caching and embedding reuse to reduce latency
• Improved retrieval relevance using hybrid search with RRF re-ranking
• Designed Postgres schema for distributed IAM (users/roles, transaction log) with constraints and replication-focused indexing
• Optimized replication hot-path query (follower catch-up by sequence number) and validated via query plans + load testing
• Built durable, replayable replication workflow with WAL, idempotent retries, and Raft-based quorum consistency
• Made explicit performance vs consistency tradeoff for sensitive identity data (strong consistency via quorum/raft)
• Built fault-tolerant distributed IAM backend in Go on AWS (Raft-based consistency)
• Debugged and fixed state divergence under network partitions via reproducible failure injection and observability
• Designed readiness gating tied to Raft commit/applied indices to prevent stale reads after recovery
• Shipped end-to-end RAG product (Qdrant + local LLM + FastAPI + React) for cross-system knowledge retrieval
• Strong hallucination-mitigation design: retrieval-first, relevance thresholds, citations, and no-answer fallback
• Designed reliable multi-step LLM agent workflow with structured planning/execution, bounded retries, and escalation/stop conditions
• Improved Postgres ETL performance using EXPLAIN-driven indexing and dbt model refactors (early filtering, incremental materialization)

Experience

Education

Awards

• Best Use of Qdrant
• Best Use of Qdrant - CodeLens hackathon
• Best Use of Operant - AWS MCP Hackathon (Jailbreak Jäger)

Skills

Languages

Publications

1 publication