s
sauhaaaard

Sauhard D

@sauhaaaard

Applied AI Engineer

Inde
Anglais
Certaines informations sont présentées en anglais.
À propos de moi
LangGraph & RAG Specialist | Production AI Agents & LLM Engineer Struggling with hallucinating LLMs, weak retrieval, or prototypes that fail in production? I build reliable, low-cost LangGraph multi-agent systems and grounded RAG pipelines that ship and scale. Live proof: SciNets.in — production platform processing 9–10 papers/run at ~$0.10 using graph-constrained multi-hop reasoning, custom eval metrics (symbolic depth, grounding stability), full stack (FastAPI, Postgres, Docker, OAuth). Tech: LangGraph, LangChain, RAG, PyTorch/Transformers, FastAPI, Docker, Python.... Plus d’infos

Compétences

s
sauhaaaard
Sauhard D
hors ligne • 

Voir mes services

Sites web IA & Logiciel
I will build grounded rag apps ai agents with langgraph langchain
De55 $US / projetPlus de détails

Expérience professionnelle

Applied AI Engineer

SciNets

Jan 2025 - Present • 1 yr 4 mos

◦ Built and deployed a multi-user agentic AI research platform for scientific literature synthesis (scinets.in) using LangGraph, RAG, FastAPI, and Postgres; processes 9–10 papers per run with ˜8–10 min latency at ˜$0.10/run. ◦ Designed a graph-constrained multi-hop reasoning engine over literature concept graphs (NetworkX), reducing spurious cross-domain links and improving evidence coherence. ◦ Implemented evidence-grounded synthesis with support/neutral/contradiction classification, grounding verdicts, and mechanistic reasoning chains. ◦ Production-hardened OAuth 2.0/JWT auth, SSE streaming, and multi-user isolation; eliminated random 401s and cross-user data leaks behind load balancers. ◦ Built real-time agent tracing UI and a custom evaluation layer measuring symbolic depth, grounding stability, and collapse rate. ◦ Diagnosed and fixed cloud-only failures across Vercel + Render; added auto-reconnect and graceful recovery for cold starts. ◦ Open-sourced research core and published related work on arXiv