Freelance AI Backend Engineer | Open Source Contributor | Production RAG

I build resilient, backend-driven AI systems. While my formal academic background is in Engineering at NIT Hamirpur, my daily execution is strictly focused on AI architecture, multi-agent orchestration, and fixing broken data ingestion pipelines.

I don't build thin LLM wrappers. I build the heavy plumbing underneath.

🏆 Top 5.58% Nationally in GATE Data Science & AI (IIT Guwahati, 2026 — 69,200+ Candidates)

• AI & Orchestration: LangGraph, LangChain, Groq, Local LLMs (Ollama), HuggingFace
• Backend & Processing: Python, FastAPI, Pandas, SQL
• Infrastructure: ChromaDB, Docker, Render
• Frontend: Streamlit

infiniflow/ragflow Core Contributor

• Google Drive Sync Engine & Memory Optimization | Merged PR #14372

  • Engineered the remote sync deletion engine for the Google Drive connector, enabling the pipeline to accurately track and reap remotely deleted files.
  • Built a memory-optimized O(N) state-reconciliation pipeline replacing heavy dictionary payloads with lightweight namedtuples, solving RAM spikes during massive enterprise snapshots while bypassing Workspace API blindspots.

• Native Docling Chunking Integration | Merged PR #14218

  • Engineered and shipped the native Docling chunking and document parsing engine into RAGFlow's main branch.
  • Resolved critical context-limit failures for enterprise pipelines by building a graceful fallback mechanism. Code reviewed and approved by core maintainers.

• Docker Build Pipeline Optimization | Merged PR #14553

  • Optimized sandbox Dockerfiles by implementing dynamic package registry fallbacks via conditional shell logic and build arguments (ARG), resolving global network timeouts during environment initialization.

Autonomous Research Agent | 🟢 Live App | ▶️ Watch Demo

• What it is: A decoupled, asynchronous research microservice currently live-deployed on Render.
• The Heavy Lifting: Built a robust Human-in-the-Loop (HITL) approval gate using LangGraph, ensuring high-fidelity outputs by maintaining strict manual oversight over the autonomous workflow. The agent independently validates scraped web data, iterates on failing searches through self-correction, and pauses to let users intercept queries before final execution. The end-to-end pipeline reliably synthesizes this verified, multi-source data into comprehensive, well-formatted PDF reports.

Zero-Leakage Enterprise RAG | ▶️ Watch Demo

• What it is: A 100% local, privacy-first pipeline that simultaneously queries unstructured PDFs and structured SQL databases.
• The Heavy Lifting: Engineered the entire stack locally using Ollama and HuggingFace embeddings to guarantee zero data leakage. Built dynamic Pandas pipelines to clean raw CSVs before SQL insertion, and wrote the routing logic to seamlessly flip between ChromaDB vector search and SQL queries based on user intent.

Currently taking on freelance contracts (15-20 hrs/wk). If your FastAPI or LangGraph backend is hitting rate limits or dropping context, I'd be happy to connect and see if I can help.

• LinkedIn
• Email: parassondhi10@gmail.com