Building a production RAG pipeline for chemical patents
Lessons from shipping semantic search over 10M+ chemical patents — chunking strategy, embedding choices, and retrieval quality.
$▋
Yogeesh Yadappanavar
AI Engineer · Bengaluru
Currently at Simreka engineering a chemical process SaaS. Interested in edge AI, biosensors, and products with real technical depth.
1.5yengineering
prodRAG / LangChain
MEbackground
experience
Software Engineer
Building Process Experiment V2 — a chemical process engineering SaaS platform for simulation, LCA analysis, and patent intelligence.
- ›Integrated RAG pipeline using LangChain + Milvus for chemical patent semantic search across 10M+ documents
- ›Architected Life Cycle Assessment (LCA) module end-to-end: 11 DB migrations, Eloquent models, repository pattern, Pinia store, and simulation callbacks
- ›Built real-time process simulation status updates using WebSockets + RabbitMQ message queuing
- ›Debugged and resolved complex Nuxt 3/Pinia session management edge cases in production
Nuxt 3Laravel 11FlaskPythonRedisRabbitMQLangChainMilvusAWSDocker
selected work
MindPattern
●activeBiosensor wearable + LLM for emotional intelligence
End-to-end platform: BLE signal acquisition from HRV/GSR sensors, edge processing, and LLM-powered cognitive/emotional state interpretation. Designed with a data moat strategy using event sourcing.
PE2 LCA Module
●shippedLife Cycle Assessment for chemical process simulation
Full-stack LCA reporting system inside Simreka's Process Experiment platform. Includes 11 DB migrations, repository pattern, Eloquent models, and Pinia store with simulation callback routing.
Chemical Patent RAG
●shippedSemantic search over 10M+ chemical patents
Production RAG pipeline using LangChain orchestration and Milvus vector database. Handles chunking, embedding, and retrieval for chemical process engineering knowledge queries.
CodeGraph + Codex CLI Workflow
●toolAI-assisted multi-language codebase navigation
Structured development workflow using CLAUDE.md context files, .claudeignore, atomic task decomposition, and session extraction prompts for navigating large TypeScript + Python codebases.
stack
AI / ML
LangChainRAG pipelinesMilvusOpenAI APIPrompt engineeringLLM fine-tuningspaCyEdge AI
Backend
PythonFastAPIFlaskLaravel 11PHPGraphQLRESTWebSockets
Frontend
Vue 3Nuxt 3TypeScriptNext.jsReactPiniaFramer Motion
Infra / Data
DockerAWSRedisRabbitMQMySQLPostgreSQLCloudflareGitHub Actions
Signal / Hardware
HRV / GSR processingBLE (Bluetooth LE)Polar H10Edge inference
Methods
Event-driven arch.CQRS / Event sourcingRepository patternCompetitive programmingDSA
writing
↗
Event-driven architecture for biosensor data: a design walkthrough
Why event sourcing makes sense when your data is physiological, time-series, and needs to power a future AI moat.
Nuxt 3 session management: the double-login bug and how I fixed it
A subtle stale watcher was setting sessionExpired after a successful refresh. Here's how I tracked it down.
contact
Let's build something.
Open to AI engineering roles — ideally working on LLM systems, RAG infrastructure, or products with real technical depth. Also happy to talk about MindPattern, biosensors, or edge AI.
Available for opportunities
Actively looking · Bengaluru / Remote
RoleAI Engineer / Full-stack
FocusLLMs · RAG · AI systems
TypeFull-time