Garv Danwani
AI Systems Engineer building real-time production infrastructure across low-latency multilingual Voice AI, computer vision physiological telemetry (rPPG), and GPU-accelerated inference pipelines that survive 2 AM production loads.
Real-Time Turn Latency Waterfall Breakdown
Multi-Tenant Real-Time Voice AI + RAG Telephony Pipeline
Enterprise voice telephony platform handling real inbound/outbound PSTN calls over Exotel IVR, streaming vector knowledge retrieval with Qdrant, and executing sub-second conversational turns with an 80%+ reduction in end-to-end latency.
rPPG Health - Contactless Physiological Telemetry Engine
AI-driven contactless physiological monitoring platform using remote photoplethysmography (rPPG) to extract blood volume pulse (BVP) waveforms directly from standard RGB webcams, with the next stage focused on CDSCO regulatory compliance.
OmniFlix - Unified Cross-Domain Recommendation System
Hybrid recommendation architecture unifying IMDb, TMDB, and MyAnimeList datasets into a multi-domain discovery engine using LightFM hybrid collaborative filtering and TF-IDF content similarity.
AI-Powered Personalized Learning & Mastery Platform
Data-driven learning recommendation engine that analyzes student performance metrics and real-time interaction telemetry to personalize study sequences and preempt knowledge drop-offs.
Financial Automation & Ledger Verification Engine
Enterprise frontend architecture and document pipeline for automated financial document processing, multi-party ledger approvals, and transaction validation.
Too many AI demos.
Not enough that survive 2 AM production.
Garv didn't start in systems engineering. He started in data science - models, notebooks, loss curves, the typical path into AI.
The deeper he went, the clearer the bottleneck became: the models that change outcomes aren't just the theoretically clever ones; they are the ones that run deterministically, sub-second, and reliably for real users under unstable network conditions.
He made a deliberate detour into backend infrastructure, WebRTC media servers, GPU inference batching, and low-level signal processing.
Today, as a founding engineer at a government-incubated healthtech venture (Project No. 21), he owns systems end-to-end: from camera sensor ingestion to preparing for the next stage in CDSCO regulatory compliance.
Avoid sequential network hops. Stream first tokens, synthesize audio in memory, and cut turn latency by 80%+.
FaceMesh 468-point dynamic ROI tracking and bandpass filters yield clinical signal stability from 30 FPS RGB webcams.
Every calculation, timestamp, and inference run must be immutable and audit-grade as next stage enters CDSCO regulatory evaluation.
AI & ML Systems Engineer (Founding Team - 1 of 5)
- ▪One of five founding team engineers at a high-velocity government-incubated health-tech startup.
- ▪Engineered contactless cardiovascular vitals pipeline using rPPG (remote photoplethysmography) to extract pulse waveforms from standard RGB cameras, achieving ~28% gain in signal stability.
- ▪Architected carrier-grade real-time Voice AI conversational pipeline streaming over Exotel PSTN telephony and WebRTC with sub-second turn latency.
- ▪Hardening the physiological telemetry pipeline and audit logs as the next stage advances toward CDSCO regulatory compliance.
Core AI/ML pipelines, signal processing, FastAPI async APIs
Next.js App Router, strict types, real-time client state
Relational modeling, query optimization, high-throughput indexes
Low-level signal compute & CUDA kernel understanding
TensorRT FastConformer ASR & low-latency neural TTS
Bidirectional Hindi ↔ English translation (1B param models)
Carrier-grade PSTN trunks, SIP gateways, 8kHz µ-law streams
Full-duplex real-time WebRTC audio streaming & room agents
Quantized Whisper models for edge/fallback transcription
Government Indian language API gateway failover
Continuous batching, PagedAttention, multi-GPU LLM serving
Ultra-fast instruction following & JSON tool calling
Dynamic model batching & concurrent model orchestration
Collection-isolated multi-tenant dense vector search & RAG
Local prototyping & GGUF quantized model evaluation
Video frame processing, color space conversions, optical flow
468-point 3D facial landmark mesh tracking & dynamic ROI
Optical blood volume pulse extraction from RGB pixels
Butterworth bandpass filters, FFT peak analysis, HRV RMSSD
Random Forest physiological proxy regression & feature scoring
Asynchronous REST endpoints, WebSockets, background task workers
Multi-tenant schemas, connection pooling, ACID audit records
In-memory caching, pub/sub event distribution, rate limiting
Containerized GPU deployments, multi-stage reproducible builds
GPU inference instances (G5/A10G), secure blob stores
Systemd services, NVIDIA driver management, shell automation
B.Tech in Computer Science & Engineering
Greater Noida Institute of Technology (GNIOT), AKTU
Specialization in Data Science & Machine Learning. Graduating First Division with deep focus on systems programming, data structures, and applied AI infrastructure.
Sarvam CODE Beta Tester
Sarvam AI EPOCH Program (2026)
Selected as specialized Beta Tester for Sarvam CODE CLI under Sarvam AI's EPOCH initiative. Executed rigorous multi-session stress testing, isolated 4 reproducible bug classes including 2 critical trust-affecting failures, and provided structured reproduction reports and fix patches.
Google Hackathon
Architected dynamic skill-to-career trajectory recommendation graphs using graph neural structures.
Smart India Hackathon (SIH 2024)
Built scalable computational solutions addressing complex real-world distributed challenges in health telemetry.
The Big Biz Theory 2025
Defended commercial viability and technical scalability for computer vision diagnostic pipelines.
Technical Design & Branding Head
Directed branding, visual communications, and technical execution for college hackathons and engineering societies.
Verified technical accreditations, research publications, competitive hackathons, and job simulations. Hover over any cell to project the full horizontal document view, or click to inspect the original PDF.
Microsoft Azure AI Essentials Workloads and Machine Learning
The Big Biz Theory 2025
Tata Job Simulation
Programming with Generative AI
IRE Journals Research Publication
Google Hackathon (Hack2Skill)
Student Development Program on DSA
Data Analytics with Python
Data Science with Python
C & C++ Systems Training
Deloitte Job Simulation
Open for Select AI Systems Roles & Engineering Consultations
Available for technical discussions regarding low-latency Voice AI pipelines, WebRTC media servers, computer vision physiological telemetry (rPPG), and production inference infrastructure.
Looking to bring me onto a project? Fill in your technical requirements, team timeline, and contact specifications to route directly to my personal email.