Meshkat Shariat Bagheri

Meshkat Shariat Bagheri

he/him

AI Engineer building local-first agents, RAG systems, OCR pipelines, and enterprise LLM applications.

Enterprise Local AI Agents at Gam Electronics

At Gam Electronics, I build local-first AI systems for enterprise workflows where privacy, permissions, retrieval quality, and operational reliability matter as much as model quality.

The work spans agentic AI platforms, protected document search, Microsoft CRM/Dynamics 365 integrations, OCR pipelines, OracleDB-backed applications, and productivity agents connected to organizational messaging and email systems.

Highlights

  • Built a fully local enterprise RAG agent platform using reasoning models, company documents, MCP/A2A-style tool catalogs, agent cards, manual LangChain tool-calling loops, streamed responses, conversation memory, and tool-call traces.
  • Integrated AI agents with RocketChat, Zulip, Microsoft CRM/Dynamics 365, Exchange/EWS, Oracle-backed internal document systems, and support-case workflows.
  • Engineered an access-controlled letter-search agent with Exchange login, Oracle email-to-user mapping, per-letter permission checks, full-text reconstruction from Milvus chunks, source previews, and feedback analytics.
  • Developed OCR pipelines with DeepSeek-OCR and other OCR/VLM models for 250k+ OracleDB document records, each containing multiple scanned pages.
  • Improved retrieval quality with dense + BM25 hybrid search, multi-concept AND ranking, adaptive score thresholds, Persian exact-match boosts, duplicate-chunk cleanup, indexed-letter discovery, and nightly incremental indexing jobs.
  • Built CRM resolved-case agents and analytics tools for semantic case retrieval, full case-detail loading, relative-date filters, service/case quality scoring, expert feedback management, aggregate statistics, topic clusters, and Shamsi metadata repair.
  • Developed productivity agents for invoice generation, meeting-minutes generation from audio, page-aware browser workflows, task import, follow-up email generation, stand-up summaries, and email RAG over indexed inbox content.

Technologies

Python, LangChain, LangGraph, Ollama, Llama.cpp, PyTorch, Transformers, FastAPI, Flask, Streamlit, Gradio, React, Milvus, FAISS, SQLite, OracleDB, Dynamics 365, OCR, RAG, and LLM fine-tuning.