8000 Soban-2004 (Soban Shankar) Β· GitHub
[go: up one dir, main page]

Skip to content
View Soban-2004's full-sized avatar

Block or report Soban-2004

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
Soban-2004/README.md
Typing SVG

AI Engineer building production-grade agentic RAG systems, multi-agent orchestration, and retrieval pipelines β€” currently open to AI Engineer roles.

Portfolio LinkedIn Email

Profile Views

GitHub Stats


πŸ“Š Impact Snapshot

Recruiter funnel narrowed Report latency cut Schema scale indexed Config effort reduced OSS PR merged
~95% fewer full rubric reviews 4.5s β†’ 1s via 7 parallel guardrail scanners 2,000+ HCM tables/views via RAG ~60–70% manual effort automated langchain-ai/langchain #39238

About Me

I'm an AI Engineer with hands-on enterprise experience through a 6-month internship at Drivestream, where I built multi-agent orchestration systems automating Oracle Fusion cloud transformation workflows β€” plus two production-grade agentic RAG platforms shipped end-to-end, from retrieval architecture to deployment.

I care about the difference between "using an LLM" and building a system that retrieves grounded evidence, reasons over it, and takes real action β€” hybrid search, reranking, guardrails, fallback chains, and evaluation, not just a prompt wrapped in a chatbot.

  • πŸ”­ Currently building agentic AI systems using LangChain, LangGraph, LlamaIndex, and MCP (Model Context Protocol)
  • 🌱 Deepening expertise in AI evaluation, prompt engineering, and production LLMOps
  • πŸ› Merged PR #39238 into langchain-ai/langchain β€” fixed a data-loss bug in RecursiveJsonSplitter
  • πŸ“ Based in Chennai, India β€” open to relocation (Bengaluru)

Featured Projects

πŸ” AI Resume & Job Matcher β€” Full-Stack RAG Platform

A dual-mode recruitment platform where job seekers get resume-vs-JD breakdowns and recruiters manage hiring end-to-end.

  • Hybrid retrieval β€” dense + sparse (BM25) fused via reciprocal rank fusion, re-scored with a cross-encoder reranker
  • Three-stage recruiter funnel narrows candidates to the top ~5%, cutting rubric reviews by ~95%
  • Three-tier LLM fallback chain (Ollama β†’ Gemini β†’ Groq) for rate-limit resilience

Next.js 16 FastAPI Supabase Qdrant

Live Demo Β· Source

πŸ€– Agentic RAG Customer Support Platform (Flipkart FAQ)

An autonomous support agent over a ~2,000-entry FAQ corpus that checks order/refund status and escalates to humans.

  • Intent-routing planner + 5 tools (FAQ search + 4 MCP tools) on LlamaIndex
  • LiteLLM gateway with fallback across 3 models / 2 providers
  • LLM Guard guardrails (7 parallel scanners) cut latency ~4.5s β†’ ~1s
  • Validated via 25-scenario RAGAS evaluation

LlamaIndex MCP LiteLLM Qdrant Docker

Live Demo Β· Source


Open Source

langchain-ai/langchain #39238 β€” Fixed a data-loss bug where RecursiveJsonSplitter.split_json() silently discarded non-dict top-level input instead of raising an error. Traced the regression to a prior fix, opened #39192, and shipped a merged fix with 9 regression tests covering lists, primitives, None, and existing dict behavior β€” validated against 149 passing tests.


Tech Stack

LLM & Agentic AI LangChain LangGraph LlamaIndex MCP OpenAI

Programming & Backend Python FastAPI Next.js SQL

Vector DB & Retrieval Qdrant Weaviate HuggingFace

LLMOps & Evaluation LiteLLM Langfuse RAGAS Groq

Cloud & DevOps Docker Azure OCI Git


πŸ“« ssoban2004@gmail.com Β· πŸ“ Chennai, India

Popular repositories Loading

  1. AI-Resume-Job-Matcher---Full-Stack-RAG-Platform AI-Resume-Job-Matcher---Full-Stack-RAG-Platform Public

    Python 1

  2. Agentic-RAG-Customer-Support-Platform-Flipkart-FAQ- Agentic-RAG-Customer-Support-Platform-Flipkart-FAQ- Public

    Jupyter Notebook 1

  3. langchain langchain Public

    Forked from langchain-ai/langchain

    The agent engineering platform.

    Python 1

  4. Freight-Cost-Watchdog Freight-Cost-Watchdog Public

    Python 1

  5. flight-pricing-prediction flight-pricing-prediction Public

    Jupyter Notebook

  6. House_price_prediction House_price_prediction Public

    Jupyter Notebook

0