CodePilot
An autonomous AI engineering system turning natural-language specs into full applications: repository understanding, architectural planning, RAG context retrieval, and self-correcting test/lint loops.
View on GitHub →
SAKSHAM PATHAK / GENERATIVE AI & VOICE SYSTEMS
I build AI systems
that talk, think & act.
Generative AI Engineer building low-latency conversational audio architectures, neural TTS adaptation, and multi-agent tool calling at Zencia AI.
TECH I WORK WITH
✦
Gemini
Multimodal & Function Calling
LangGraph
Stateful Agent Loops
Voice AI
STT → LLM → TTS Streaming
RAG
Retrieve → Rerank → Generate
FastAPI
Async WebSockets & Endpoints
GCP
Cloud Run & Serverless
MCP
Model Context Protocol
•••
Same
Person
Different
Possibilities.
Let's build
a more open AI future.
Building at
Zencia AI →
Voice AI
STT • VAD • Streaming TTS
Agentic Systems
Plan → Tool → Result
LLM Applications
Prompt → Reason → Output
Production Scale
99.9% Uptime • <50ms
Silero VAD • 24kHz Stream
DUPLEX LIVE
VAD State
SPEECH ACTIVE
Agent Stream
STREAMING (24kHz)
Latency: <22ms
Career trajectory / 5 Chapters
Five chapters, one thread.
From scaling devotional content to a 100K+ audience, to pure mathematics, to evaluating frontier LLMs, to building production voice AI at Zencia AI—the thread has always been building systems that truly work for people.
CASE / 01
2021 – 2024
Creative Genesis
While pursuing B.Sc. Mathematics at Dr. Ram Manohar Lohia Avadh University, I built a 100K+ following on Instagram and YouTube editing devotional content—an early lesson in what makes ideas resonate at scale.
CASE / 02
Aug 2024
The Pivot
Began my M.Sc. in Artificial Intelligence & Machine Learning at IIIT Lucknow, channeling rigorous mathematical foundations into transformer architectures and AI systems engineering.
CASE / 03
Oct 2024 – Mar 2025
Outlier
Worked as a freelance AI developer focused on LLM evaluation, prompt engineering, and response-quality benchmarking—dissecting how frontier models reason under the hood.
CASE / 04
2025
Shipping in Public
Architected and shipped FALCON, DocuMind AI, and parth-dl under the Parthmax handle—releasing open-source tools on GitHub, Hugging Face, and PyPI.
CASE / 05
Aug 2025 – Present
Zencia AI
Generative AI Engineer building real-time conversational voice agents, speech models, and agentic workflows in production—graduating with M.Sc. in Jun 2026 alongside the role.
Technical arsenal / Full stack
Everything I build with.
Not a wish list—this is the battle-tested stack behind FALCON, DocuMind AI, CodePilot, and the production voice pipelines running at Zencia AI.
Languages
LLM Engineering
Models
LLM Infra
RAG & Retrieval
Voice AI
Backend
Deployment
Deep Learning
Field notes / Production work
What I'm building at Zencia AI.
Since August 2025, I have been engineering production conversational voice AI and autonomous agentic systems at Zencia AI.
01
Real-Time Voice Pipelines
Low-latency STT (Whisper, Deepgram), LLM reasoning, and streaming TTS over WebRTC & Twilio—with barge-in detection and Voice Activity Detection for natural conversation.
02
Neural TTS Adaptation
Adapting transformer architectures for expressive speech synthesis by bridging text token embeddings with audio representations.
03
Agentic AI & Tool Calling
LangGraph multi-agent architectures equipped with tool calling and structured outputs, integrated with Google Calendar, CRMs, and live databases.
04
Production Microservices
High-throughput async FastAPI services, WebSockets, background tasks, structured telemetry—containerized with Docker on GCP Cloud Run.
32Public GitHub repos
67GitHub followers
11Hugging Face Spaces
1Package on PyPI
Field evidence / Shipped work
From idea to shipped code.
Public, reproducible projects built under the Parthmax name—live on GitHub, Hugging Face, and PyPI.
FALCON
A multi-stage automated fact-checking pipeline: claim extraction, cross-referencing Wikipedia & live web sources, LLM verification, and explainable confidence scoring.
Try on Hugging Face →DocuMind AI
Document intelligence over large PDF collections—BGE embeddings, FAISS/ChromaDB retrieval, BM25 & semantic reranking, orchestrated across Gemini, Claude, and open-source models.
View on GitHub →parth-dl
A lightweight, zero-dependency Instagram media downloader CLI—no login or API keys required. Engineered and maintained in pure Python.
View on PyPI →movie-box
A Python terminal CLI to search, stream, and download media—featuring a cyberpunk interactive shell, async API queries, subtitle handling, and MPV/VLC streaming integration.
View on GitHub →100 Best GenAI Projects 2025
A curated repository of 100+ high-impact open-source LLM, RAG, and AI agent projects—actively maintained as the state of the art progresses.
View on GitHub →Credentials / Track record
Proof, not just claims.
Formal academic credentials, algorithmic problem-solving, and verified competencies backing my applied AI engineering work.
Algorithms
LeetCode DSA
300+ Data Structures & Algorithms problems solved, providing the algorithmic foundation behind efficient production system architecture.
View LeetCode Profile →Certifications
AI Agents & Systems
- AI Agents — Hugging Face
- Advanced Model Context Protocol — Anthropic
- Data Analytics — Deloitte Australia
Education
IIIT Lucknow
M.Sc. Artificial Intelligence & Machine Learning (Aug 2024 – Jun 2026).
B.Sc. Mathematics, Dr. Ram Manohar Lohia Avadh University (2021 – 2024).
Get in touch / Let's collaborate
Let's build something exceptional.
Open to discussions around voice AI, agentic LLM systems, engineering collaborations, and cutting-edge GenAI challenges.