> hello_world

Darsh Patil

AI Engineer | Building Voice AI, Agentic AI Systems & Full-Stack AI Products

Hey, I'm Darsh.

I build real-world AI products — from real-time voice agents and agentic workflows to scalable full-stack applications. I enjoy working at the intersection of LLMs, product engineering, system design, and automation. Currently exploring opportunities where I can build impactful AI products at scale.

Building AI systems that actually solve problems — not just demos.
Darsh Patil
> about_me

About Me

I'm an AI Engineer and Full Stack Developer who enjoys building intelligent systems that feel practical, fast, and useful.

I like building complete products end-to-end rather than isolated demos. I've worked on production CRM systems, payment integrations, scheduling systems, collaborative real-time apps, and AI-native products.

What excites me most is building AI that interacts naturally — voice conversations, autonomous workflows, retrieval systems, and real-time experiences.

My work spans

Voice AI pipelinesLLM orchestrationAgentic AI workflowsReal-time AI systemsBackend architectureFrontend product development

Especially interested in

Voice AIAI AgentsAgentic workflowsRAG systemsAI infrastructureScalable product engineeringApplied AI products
0+

Production AI Systems Built

0+

Full Stack Products

0+

Real-Time Applications

0+

AI Experiments / Projects

> how_i_build

How I Build Things

Not just a list of projects — the thinking behind each one. The problem, the constraints, the architecture, and what actually came out the other side.

> projects

Stuff I've Built

SV
Voice AI

Self-Improving Voice AI Sales Agent

Full-duplex multilingual voice sales agent (Silero VAD → Sarvam ASR → Gemini → streaming Sarvam TTS) grounded on gated Qdrant RAG — with an offline "Dream Engine" that self-improves the bot from failed calls when no one is on the line, auto-injecting approved prompt fixes into the next conversation. Barge-in, 11 Indian languages with mid-call switching, and per-provider token isolation.

PipecatGemini Flash-LiteGroq gpt-oss-120bSarvam ASR/TTSSilero VADLangGraphQdrantFastAPI
Code
L–
Agentic AI

LawAI – Agentic Legal RAG

4-way hybrid retrieval RAG (BM25 + semantic + exact phrase + metadata) with RRF reranking for Indian legal documents. Source citations and verification mode.

LangChainPineconeOllamallama3BM25Streamlit
Code
T
Agentic AIIn Development

TalentLens

Adaptive voice interview platform with Gemini-powered questioning, Sarvam STT/TTS over WebSocket, sandboxed code execution, multi-layer proctoring, and BYOK key encryption.

FastAPIGoogle ADKGeminiSarvam AIWebSocketReactPostgreSQL
S
Agentic AIIn Development

SecondSelf.ai

Autonomous AI meeting agent that joins calls as your representative — live transcription, voice conversation, RAG-grounded memory, and post-meeting action items pushed to Notion.

PipecatLangGraphAttendeeDeepgramCartesiaQdrantMem0Notion API
> skills

Tech I Work With

AI / LLM Engineering

LangGraphLangChainRAGAgentic AILLM OrchestrationPrompt EngineeringVoice AI (ASR / TTS)Vector RetrievalEmbeddingsTool Calling

Backend & APIs

PythonFastAPIWebSocketsREST APIs

Data & Cloud Infra

PostgreSQLVector DBs (Qdrant / Pinecone)DockerAWSGCP

Full-Stack

ReactNext.jsTypeScript

System Design

RAG System ArchitectureAgentic Pipeline DesignReal-Time Voice PipelinesMulti-Agent OrchestrationVector Memory SystemsAsync Worker Queues
> contact

Let's Build Something Cool

If you're building something in AI, voice interfaces, automation, or product engineering — I'd love to chat.

Email MeLinkedInGitHubResume