Skip to content
View samarth70's full-sized avatar
4 followers · 30 following

Achievements

Achievement: Pull Shark

Achievements

Achievement: Pull Shark

Block or report samarth70

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
samarth70/README.md

Hi, I'm Sam Agarwal 👋

AI Software Engineer — production LLM systems, multi-agent architectures & edge-native AI
MS Data Science @ Stony Brook University · ex-Genpact (2.5+ yrs) · ex-ARMA AI Labs, CA, US

🔭 What I build

I turn LLMs into reliable, cheap, production systems : RAG pipelines, agentic workflows, LoRA fine-tuning, and evaluation harnesses that make model behavior measurable. Recently focused on edge-native AI (Cloudflare Workers, Pyodide/WASM) that runs fast and costs near-zero.

🚀 Featured projects

Project What it does Stack
ParamMem Parametric reflective memory for LLM agents: cut agent reflection latency 94% (45s→2.8s), +11.4% HumanEval LoRA · Llama-3 · Groq · ChromaDB · FastAPI · React
PrepGenie Edge-native AI interview coach + resume reviewer; recursive LLM failover, LLM-as-Judge scoring Pyodide/WASM · Gemini→Groq→Mistral · Gradio
AI Pentest Scanner Autonomous multi-agent security scanner: recon → vuln detection → auto-report Multi-agent · Groq/Gemini · Nmap/Shodan · React/Flask
Real-Time Stocks (Serverless) Global stock dashboard on the edge; pull-on-demand caching → 1 API call/min/symbol at scale Cloudflare Workers · D1 · R2 · Vite

🤗 Also on Hugging Face — 15+ live Spaces including Neural Style Transfer (62 ❤️), plus PrepGenie, the AI Pentest API, and the ParamMem reasoning dashboard.

🌱 Open source

I contribute fixes to the backend and AI tooling I use day to day — a few of the most significant ones in Supabase (auth & storage), plus ongoing work across agent-evaluation and MCP projects.

🛠️ Tech

AI/LLM: PyTorch · TensorFlow · LangChain · RAG · LoRA · MCP · DeepEval · Agentic Workflows · LLM-as-Judge Backend: Python · Java/Spring Boot · C++ · FastAPI · Flask · REST · SQL Data & Infra: ChromaDB · Pinecone · Pandas · Spark · AWS · Cloudflare · Docker · Git

💼 Experience

  • Generative AI Engineer Intern — ARMA AI Labs (2025): built a DeepEval-based eval framework (7 metrics) + a production model-routing pipeline.
  • Senior Associate Software Developer — Genpact (2022–2024): shipped Salesforce/Spring Boot systems for High Profile Clients; cut manual data entry 60%.

Open to AI/ML Engineer roles · always happy to collaborate on LLM & agent projects.

Pinned Loading

  1. AI-Pentest-Scanner AI-Pentest-Scanner Public

    Autonomous multi-agent AI penetration-testing tool — recon, vulnerability detection and auto-reporting via Groq/Gemini agents with Nmap/Shodan. React + Flask + Cloudflare.

    Python 1

  2. E-Election-Assist-Project/Manifesto-Explainer E-Election-Assist-Project/Manifesto-Explainer Public

    Analyzing Political Parties Manifesto With Text Mining. Part Of My B.Tech Final Year Project.

    Python 1

  3. ParamMem ParamMem Public

    Parametric reflective memory for LLM agents — a LoRA-tuned memory module that cuts agent reflection latency 94% (Llama-3, Groq, ChromaDB, FastAPI + React dashboard).

    Python 1

  4. Prepgenie-Mock-Interview-system Prepgenie-Mock-Interview-system Public

    Edge-native AI mock-interview coach and resume reviewer — Pyodide/WASM, recursive LLM failover (Gemini→Groq→Mistral) and LLM-as-Judge scoring.

    Python 1

  5. Political-Party-Symbol-Detector Political-Party-Symbol-Detector Public

    Custom YOLOv5 detector for political party symbols, with a deployed inference API.

    Python 2

  6. real-time-stocks-serverless real-time-stocks-serverless Public

    Real-time stock dashboard on the Cloudflare edge (Workers + D1 + R2) with pull-on-demand caching — one upstream API call per minute per symbol at scale.

    JavaScript 1