Skip to content
View AasthaPJoshi's full-sized avatar
4 followers · 11 following

Block or report AasthaPJoshi

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.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. 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
AasthaPJoshi/README.md
Screenshot 2026-07-14 at 1 22 28 AM

LinkedIn Email GitHub


🛠 What I've Shipped

Every project below has a quantified eval score, an observability layer, and a deployed URL. Most intensive pipelines live in private repos due to regulatory data constraints — the code is real, the metrics are real.

🔐 ThirdLine-of-Defense

Agentic AI Audit Platform for Financial Compliance

Regulatory AI that actually works under real compliance constraints (SR 26-2 · OCC Bulletin 2026-13)

  • 5-agent LangGraph orchestrator · SHA-256 hash-chained audit ledger
  • Human-in-the-loop governance gate · FastAPI + React dark dashboard
  • Targets financial services governance — not a tutorial, a real use case

Result: F1 = 0.909 · Precision = 0.923 · Recall = 0.895

Python LangGraph FastAPI React PostgreSQL

Repo

🏥 MediSight+

Full-Stack Clinical AI SaaS Platform

Production clinical AI with role-based access, multi-model routing, and eval harness

  • 3 role-based portals: Patient · Doctor · Billing
  • LangGraph RAG pipeline · GPT-4o-mini / GPT-4o / Claude Sonnet routing
  • Ragas evaluation harness · Deployed on Vercel + Railway

Result: Faithfulness = 0.93 · Answer Relevancy = 0.91

Python LangGraph OpenAI Vercel Railway

Repo Live

🔬 RetrievalLab

Cross-Industry RAG Benchmarking Platform

The platform I built because I couldn't find a rigorous RAG benchmarking tool that actually worked

  • 5-node LangGraph pipeline · Hybrid retrieval: dense + BM25 + RRF + Cohere rerank
  • ObserveLab observability layer · 19-document clinical corpus
  • React 18 + TypeScript frontend · 35-page technical documentation

Result: NDCG@10 = 0.847 · MRR = 0.891 · Confidence gating implemented

Python LangGraph pgvector React TypeScript

Repo

💰 FinSight360

Autonomous Financial Risk Intelligence via SEC EDGAR

AI-powered financial analyst that reads SEC filings so you don't have to

  • Autonomous SEC EDGAR extraction · Multi-document RAG pipeline
  • Risk scoring engine · Financial ratio analysis + trend detection
  • Deployed on Streamlit Cloud — live and running

Result: Processes 10-K filings end-to-end · Deployed to production

Python Streamlit LangChain SEC EDGAR

Repo Live

🤖 Dual-Mode Conversational AI Assistant

Multi-Model Chat System with Mode Switching

Because one LLM is never enough — built a system that routes intelligently between models

  • Dual-mode architecture: RAG mode + direct LLM mode with smart switching
  • Multi-model support · Conversation memory · Context-aware routing
  • Clean UI with mode toggle and source attribution

Result: Seamless model switching · Production-grade conversation management

Python LangChain OpenAI Streamlit

Repo

📝 MCP CLI Note App

Model Context Protocol Note-Taking CLI

Built when MCP was brand new — because I had to know how it worked from the inside

  • Full MCP server + client implementation from scratch
  • CLI-native note creation, retrieval, search, and tagging
  • Clean protocol design · Tool schema · Real MCP handshake

Result: End-to-end MCP implementation · Protocol-level AI tool experience

Python MCP CLI

Repo

🧠 Stack — By Layer

Aastha Joshi Skill Depth Map

📊 GitHub Activity

Actively building since Jan 2025 · 2 publications · 3 production-deployed AI systems · 6 public repos Most intensive work (LangGraph pipelines, eval harnesses, compliance infra) lives in private repos due to data sensitivity.

🤝 Let's Connect

💡 Collaborate With Me

Have an idea? A half-built product? A problem that needs an AI system?

Research · Freelance · Full-time · Open source — all on the table.

If you're building something real, let's talk.


Collaborate on LinkedIn

🚀 Hire Me

Open to Applied AI Engineer, Forward Deployed Engineer, LLM Engineer, Data Analyst, and AI PM roles in the US.

I bring production AI, research depth, and business thinking to the same table. If that's what your team needs — let's talk.


Hire Me on LinkedIn

🌐 Find Me

LinkedIn GitHub Email

Visitor Count

MS Information Systems · SDSU Fowler College of Business · GPA 3.73 · San Diego, CA · Open to Work 🇺🇸

Pinned Loading

  1. MediSight-Plus MediSight-Plus Public

    Clinical AI platform: LangGraph RAG over PubMed, ICD-10/CPT/HCPCS billing intelligence, 3 role-based portals. FastAPI · PostgreSQL · Pinecone · React · AWS EC2 · Faithfulness 0.93

    Python 1

  2. ThirdLine-of-Defense ThirdLine-of-Defense Public

    ThirdLine is a production-grade agentic AI audit platform. It is an AI system that autonomously audits other AI systems.

    Python 1

  3. Dual-Mode-Conversational-AI-Assistant Dual-Mode-Conversational-AI-Assistant Public

    Open-source dual-mode AI assistant for chat and voice, featuring conversational intelligence, multimodal interaction, and productivity-focused automation.

    Python 2

  4. FinSight360 FinSight360 Public

    Autonomous financial risk intelligence platform, SEC EDGAR ingestion, Isolation Forest anomaly detection, Benford's Law forensic analysis, corporate network graph, SHAP explainability, and Gemini A…

    Python 1

  5. MCP-CLI-Note-App MCP-CLI-Note-App Public

    A CLI assistant built with the Model Context Protocol (MCP) featuring FastMCP tools, resources, prompts, agentic tool loops, and a prompt-toolkit shell with @doc mentions and /slash commands.

    Python 1

  6. RetrievalLab RetrievalLab Public

    Cross-industry RAG benchmarking platform, 10 chunking strategies, 3 retrieval modes (BM25/Vector/Hybrid RRF), 5-node LangGraph agent, NDCG@K · MRR · MAP · Ragas · adversarial robustness evaluation …

    Python 1