Sepehr Radmard

Self-hosted multi-model AI workspace with side-by-side arenas, judged debates, a workflow canvas, sandboxed code execution and a DLP gate.

• FastAPI
• React
• OpenRouter

Code for polymind on GitHub Live demo of polymind

Local-first AI copilot for project managers: MCP tools across Jira, GitHub, Slack and more, approval-gated writes, natural-language rules engine.

• MCP
• FastAPI
• React

Code for pm-assistant on GitHub

Talk to a chess coach and play by voice in Persian or English: LiveKit voice agent + tool-calling model driving a live board, python-chess owns the rules.

• LiveKit
• React
• python-chess

Code for voice-chess-coach on GitHub

Paste a voice agent's prompt, get a judged test suite back: LLM-generated scenarios run against a LiveKit agent, scored turn by turn by an LLM judge.

• LiveKit
• LLM-as-judge

Code for voice-agent-testgen on GitHub

Local, Persian-first AI workbench where agents call native tools and MCP servers, and every step shows up in the chat.

• Next.js
• AI SDK
• MCP

Code for agent-studio on GitHub

Fly a 6DOF probe through a real-scale solar system in the browser, every planet placed from NASA/JPL Horizons data.

• Three.js
• WebGPU
• Vite

Code for true-anomaly on GitHub

Steel price report PDFs in; charts, spreads and a grounded AI market read out, with every number traced to a deterministic parser.

• FastAPI
• React
• Multi-agent

Code for steel-market-analyst on GitHub

Full-screen terminal coding agent on Bun + OpenTUI: reads, edits, searches and runs code with hard step and dollar caps per turn.

• Bun
• OpenTUI
• OpenRouter

Code for sepicode on GitHub

Hybrid retrieval API for LLM agents: multi-query fusion, dense + sparse search over LlamaCloud vector indexes, Cohere rerank.

• FastAPI
• LlamaIndex
• Cohere

Code for rag-service on GitHub

Find where your RAG chatbot is wrong: LLM-as-judge on four criteria plus side-by-side human expert scoring.

• Flask
• LLM-as-judge

Code for rag-evaluator on GitHub Live demo of rag-evaluator

Product Talk: scan a supermarket product and ask it out loud about protein, sugar or allergens. Realtime voice agent grounded in pack facts.

• LiveKit
• OpenAI Realtime
• Next.js

Code for product-assistant on GitHub

Seven AI reviewers on every GitLab merge request, one human who decides what gets posted. Multi-agent code review with a CTO dashboard.

• n8n
• GitLab
• Next.js

Code for pr-agent on GitHub

Persian AI phone receptionist: answers a real call over Asterisk AudioSocket, STT → LLM → TTS in natural Persian, with an Android companion app.

• Asterisk
• STT/TTS
• Android

Code for phone-agent on GitHub

Persian meeting recorder: chunked speaker diarization with speaker stitching, schema-validated LLM summaries, searchable minutes.

• FastAPI
• Next.js
• Diarization

Code for meeting-assistant on GitHub

Persian speech-to-speech voice agent on LiveKit that answers only from retrieved passages (RAG), built for filtered networks.

• LiveKit
• RAG
• Realtime

Code for livekit-fa-agent on GitHub

Extract invoice fields with OCR + an LLM, then check every value against the exact box it came from. Human-in-the-loop review UI, English and Persian.

• PaddleOCR
• FastAPI
• Structured output

Code for invoice-extractor on GitHub

Plain-Persian bug reports in, structured AI drafts out: vision-LLM drafting, duplicate detection and truncated-JSON recovery.

• Next.js
• FastAPI
• Vision LLM

Code for feedo on GitHub

Ask your ledger questions in Persian, get audited numbers back: read-only accounting dashboard plus an LLM tool-calling SQL agent over an ERP warehouse.

• Text-to-SQL
• Hasura
• FastAPI

Code for erp-sql-agent on GitHub

Persian bill splitter: type the expense in Farsi, an LLM drafts it via tool calls, code does the math (min-transfer settlement, tested).

• Next.js
• Drizzle
• Vitest

Code for dongeto on GitHub

Ask any question, get an interactive HTML/SVG explainer page planned and designed by a multi-stage LLM pipeline and streamed into a sandboxed iframe.

• Next.js
• AI SDK
• Generative UI

Code for ai-explain on GitHub Live demo of ai-explain

Realtime voice assistant with a live Tavus virtual avatar in the room: OpenAI Realtime, MediaPipe face detection triggers, LiveKit tokens.

• LiveKit
• OpenAI Realtime
• Tavus

Code for live-media-tavus on GitHub