| Focus | What it looks like in code |
|---|---|
| Agentic workflows | Multi-agent pipelines with LangGraph — supervisors, specialized workers, resumable graphs |
| Real-time voice & avatars | WebRTC agents with Deepgram STT, LLM reasoning, TTS, and Simli lip-sync |
| Multimodal media | Video intelligence, narrated audio programs, SDXL thumbnails, episodic story generation |
| Retrieval systems | Domain RAG with Chroma, citations, confidence scoring, and tool-augmented search |
| Applied product AI | SEM strategy generators, market/ads agents, interview systems, LMS tooling |
Real-time conversational AI
Simli · Deepgram · LiveKit stacks — Enterprise-style avatar conversation apps with Next.js clients, streaming STT/LLM/TTS, and lip-synced video personas.
TypeScript Python Pipecat WebRTC Deepgram Simli
Also: simli-livekit-frontend · simli-livekit-backend
Pipecat Avatar Backend — Low-latency pipeline: WebRTC → Deepgram → LLM → Cartesia/XTTS → Simli, with Silero VAD for Ambilio-style avatar experiences.
Python Pipecat OpenAI Cartesia
Agentic systems & RAG
legal_assistant ★ — Agentic legal RAG with LangGraph + Gemini + Tavily + Chroma: query routing, citations, confidence scores, Streamlit research UI.
Python LangGraph LangChain Gemini Chroma
multi-ai-dev-system — Full SDLC automation: BRD → architecture → code → tests → docs via temperature-tuned agents and a Vue monitoring UI.
Python LangGraph FastAPI Vue RAG
Stock Analyzer — Multi-agent market signal aggregator: price, news, and sentiment agents into trading signals.
Python LangGraph FastAPI React
AI SEM Planner — Search / Shopping / Performance Max strategy generator with keyword discovery, scoring, and exportable deliverables.
Python Streamlit Agents
Multimodal creative pipelines
Ai_video_suite — Video intelligence: VideoDB transcription, semantic search, meeting summaries, clip generation, multilanguage dubbing.
Python React LangGraph VideoDB ElevenLabs
Audio Generation Pipeline — Script/audio → narration → mood-aware BGM → mixed program output (CLI + Streamlit).
Python Gemini ElevenLabs Streamlit
Story Generation — Episodic AI storytelling with persistent character/plot memory and AI critique.
Python Memory Vector store Streamlit
Image Generation · SDXL ★ — LoRA fine-tuning + Gemini-guided YouTube thumbnail generation from title and concept notes.
Python SDXL LoRA Gemini
Beyond software — systems thinking
MSMD — Graph-based synthesis for multi-state mechanical devices: topology search, DOF (Grübler), building-block knowledge bases, and transformation rules.
| Layer | Tools |
|---|---|
| Agents | LangGraph · LangChain · Deep Agents · Pipecat |
| Models | Gemini · OpenAI · local / edge TTS & vision |
| Realtime | WebRTC · LiveKit · Deepgram · Simli · Cartesia · ElevenLabs |
| Data / RAG | Chroma · VideoDB · Tavily · structured memory stores |
| Apps | FastAPI · Next.js · React · Streamlit · Vue |
| Infra | Docker · uv · Render · Prometheus-style metrics |
- Production-grade avatar agents (Simli + LiveKit / Pipecat) with tight STT→LLM→TTS loops
- Domain RAG that cites sources and routes between corpora vs. live web search
- End-to-end media generation systems — video, audio, thumbnails, long-form narrative
- Multi-agent product automation for ads, SEM, interviews, and learning experiences
Open to collaborations on agentic products, multimodal pipelines, and real-time AI interfaces.
Repositories · legal_assistant · Ai_video_suite · Simli-Deepgram-Agent