Building production-grade AI systems backed by strong enterprise software engineering.
I’m an AI Engineer and enterprise software engineer with 4+ years of experience building production systems, automation platforms, backend services, data workflows, and enterprise integrations.
My current focus is on building practical AI systems across:
- Generative AI & Large Language Models
- Retrieval-Augmented Generation (RAG)
- Agentic AI & Multi-Agent Systems
- Natural Language Processing
- Vector Search & Semantic Retrieval
- Multimodal AI
- LLM Evaluation & Observability
- Production AI APIs and MLOps
At Asian Paints, I work on business-critical engineering systems across manufacturing, supply chain, R&D, sourcing, and enterprise operations. That background gives me a strong foundation in reliability, integration, data engineering, production support, and measurable business impact.
dinesh = {
"role": "AI Engineer",
"experience": "4+ years",
"focus": [
"Generative AI",
"RAG",
"Agentic AI",
"LLM Applications",
"NLP",
"Multimodal AI",
"Production AI Engineering",
],
"core_stack": [
"Python",
"FastAPI",
"PyTorch",
"LangChain",
"PostgreSQL",
"pgvector",
"FAISS",
"Docker",
],
"enterprise_depth": [
"Java",
"Spring Boot",
"Oracle SQL / PL/SQL",
"SAP S/4HANA",
"Infor Optiva PLM",
"ETL & Integrations",
],
}Generative AI
LLMs
RAG
Agentic AI
Multi-Agent Systems
Prompt Engineering
NLP
Embeddings
Semantic Search
Vector Databases
LLM Evaluation
Tool Calling
LangChain
LangGraph
LlamaIndex
Hugging Face
Sentence Transformers
Ollama
OpenAI API
Gemini API
FAISS
pgvector
HNSW
Hybrid Search
Reranking
Reciprocal Rank Fusion
Technologies I am actively implementing and strengthening through projects.
MLflow
MLOps
LLMOps
Kubernetes
CI/CD
AWS
Azure
GCP
PEFT
LoRA
QLoRA
Model Monitoring
Model Serving
Fine-Tuning
A production-focused RAG system built around grounded answers, hybrid retrieval, citations, evaluation, and observability.
Stack: Python · FastAPI · PostgreSQL · pgvector · Ollama · Sentence Transformers · OpenTelemetry
- Supports PDF, Markdown, web pages, Notion, OCR, and voice-note ingestion
- Combines pgvector HNSW dense retrieval + PostgreSQL Full-Text Search
- Uses Reciprocal Rank Fusion (RRF) for hybrid retrieval
- Implements citation validation, evidence gates, context budgeting, and refusal behavior
- Evaluates retrieval and generation using:
- Recall@k
- MRR
- nDCG
- Faithfulness
- Citation quality
- Refusal quality
- Includes observability with OpenTelemetry
Focus: RAG · Hybrid Retrieval · Vector Search · LLM Evaluation · Production AI
Multi-agent supply-chain intelligence system combining live disruption signals with enterprise data.
3rd Place — Asian Paints Hackathon
Stack: LLMs · Agentic AI · Multi-Agent Systems · Microsoft Copilot Studio · NLP
Designed specialized agents for:
- News Intelligence — analyzes trusted global and local news
- Route Risk — evaluates logistics and supplier-route disruption
- Alternate Raw Material Recommendation — identifies alternatives
- Call-to-Action — produces mitigation recommendations
The system connects external disruption signals with enterprise master data to generate:
- Explainable supplier and raw-material risk scores
- Affected finished-goods insights
- Alternate material recommendations
- Potential savings and mitigation actions
Focus: Agentic AI · Multi-Agent Orchestration · Enterprise AI · Supply Chain Intelligence
NLP and semantic-matching platform for evaluating resume-to-job fit.
Stack: Python · FastAPI · Sentence Transformers · spaCy · TF-IDF · Gemini API · Oracle · Docker
- Parses resumes with multi-format support and OCR fallback
- Generates semantic embeddings for resume-to-job matching
- Implements a 6-component composite scoring pipeline
- Extracts skills and identifies:
- Critical gaps
- Recommended gaps
- Optional gaps
- Generates AI-assisted resume improvement suggestions
- Exposes the scoring workflow through FastAPI
- Containerized with Docker
Focus: NLP · Embeddings · Semantic Similarity · FastAPI · Applied AI
Privacy-first semantic search for local photo and video libraries.
Stack: Python · PyTorch · OpenCLIP · FAISS · OpenCV · SFace · SQLite
- Natural-language text-to-image search
- Reverse image-to-image search
- OpenCLIP / PyTorch multimodal embeddings
- Persistent FAISS nearest-neighbor index
- Batched GPU/CPU embedding generation
- Incremental media indexing
- Representative video-frame search
- Face detection, grouping, naming, and merging
- Virtualized rendering designed for 10,000+ media results
- Entirely local-first for privacy
Focus: Multimodal AI · Computer Vision · CLIP · Vector Search · GPU Inference
Architecture-aware context layer for AI coding agents.
Stack: TypeScript · Node.js · Model Context Protocol · Tree-sitter · SQLite · FTS5
CodeAtlas converts a repository into an incremental, provenance-aware code graph that AI agents can query instead of blindly searching source files.
It indexes:
- Files and symbols
- Function and method calls
- Routes and API relationships
- Database models
- Dependencies
- Architecture relationships
It exposes bounded MCP tools for:
- Code search
- Execution traces
- Impact analysis
- Architecture exploration
- Repository health
- Evidence-backed source context
Focus: MCP · AI Agents · Code Intelligence · Static Analysis · Developer Tooling
Before specializing deeply in AI engineering, I built and supported production systems where correctness, reliability, traceability, and integration quality are non-negotiable.
- Built and supported 20+ production systems
- Systems used by 200+ enterprise users
- Reduced PLM-SAP synchronization errors and incidents by approximately 40%
- Maintained 95%+ release success
- Automated workflows eliminating 8+ hours of daily manual effort
- Reduced selected processing workflows from hours to seconds
- Automated base/hardener code creation from 40–50 minutes to seconds
Java
Spring Boot
Oracle SQL
PL/SQL
REST APIs
SAP S/4HANA
SAP RFC/BAPI
Infor Optiva PLM
Informatica ETL
FileNet
Batch Processing
Enterprise Integrations
NOW NEXT DEEPER
───────────────────────────── ────────────────────────────── ─────────────────────────────
Production RAG systems → LangGraph agent workflows → Production Agentic AI
Hybrid retrieval MLOps / LLMOps Multi-agent orchestration
LLM evaluation MLflow & observability Model lifecycle engineering
Local multimodal AI Kubernetes deployment Scalable inference
FastAPI AI services AWS / Azure / GCP AI platform engineering
My goal is to work on production AI systems where model capability and strong software engineering meet.
I'm interested in conversations around:
AI Engineering
Generative AI
RAG
Agentic AI
LLM Applications
AI Developer Tools
Enterprise AI
Open Source