Sahil Tomar (dev-S-t)

What does Sahil Tomar work on?

Sahil Tomar (dev-S-t) works on production Voice AI infrastructure, LiveKit/WebRTC telephony bridges, multi-tenant RAG systems, and agentic orchestration platforms using Google ADK, LangGraph, FastAPI, Python, and GCP.

What is dev-S-t known for?

dev-S-t is the online entity handle for Sahil Tomar. He is known for high-throughput Voice AI SaaS development (1,000+ calls/day, sub-200ms p95 latency), IEEE-published research on software-driven supply chain optimization, and private NAT gateway telephony deployments.

What voice AI infrastructure has Sahil Tomar built?

Sahil Tomar engineered VOAG (enterprise Voice AI SaaS handling 1,000+ daily calls), deployed a UAE SIP/WebRTC telephony bridge inside a private VM behind strict NAT gateways, and led the Hireups rescue engagement migrating a WebRTC monolith to GCP microservices with LiveKit Simulcast/Dynacast optimization.

AI Solutions Engineer — TechieMaya

Core Focus: Enterprise AI solutions, Voice AI pipelines, RAG architectures, client deployments.

• Engineered production Voice AI infrastructure (VOAG) handling 1,000+ calls per day at sub-200ms p95 latency.
• Deployed UAE telephony bridge inside partner VM environment behind strict NAT gateways.
• Architected agentic orchestration pipelines with Google ADK, LangGraph, tenant-aware tool registries, and OAuth 2.0 authentication.

Freelance Voice AI Consultant — Quantashift / MGS Technology (Hireups Engagement)

Core Focus: System rescue, WebRTC streaming optimization, GCP microservices migration.

• Migrated failing WebRTC monolith to modular GCP Cloud Run and Compute Engine microservices.
• Implemented LiveKit Simulcast and Dynacast for dynamic audio/video bandwidth adaptation.
• Configured multi-LLM fallback routing for continuous uptime and integrated browser-side CV/NLP inference.

Machine Learning Intern — Infosys Springboard

Core Focus: Machine learning data pipelines, model optimization, predictive analytics.

• Constructed predictive machine learning workflows, evaluating model accuracy and data preprocessing strategies.

VOAG — Enterprise Voice AI SaaS

URL: /projects/voag/ | .md Mirror

High-throughput Voice AI platform serving 1,000+ daily calls with sub-200ms p95 latency. Features UAE telephony bridge deployed behind private NAT gateways, streaming STT/TTS audio pipelines, and WebRTC/SIP integration.

Audalimo — WhatsApp AI Dispatcher

URL: /projects/audalimo/ | .md Mirror

Automated AI dispatcher running on WhatsApp. Designed with an asynchronous gateway architecture to bypass Meta WhatsApp API rate limits and concurrency bottlenecks.

Privacy-First On-Prem RAG

URL: /projects/privacy-rag/ | .md Mirror

Local RAG pipeline built for sensitive data security. Utilizes on-premise vector storage and LightRAG hybrid search to guarantee zero external data leakage.

AnyAssist — Multi-Tenant RAG Platform

URL: /projects/anyassist/ | .md Mirror

Multi-tenant retrieval platform with vector database namespace isolation, hybrid keyword and vector retrieval, and LiteLLM model routing for cost optimization.

UniBias — Live Attention Tracker

URL: /projects/unibias/ | .md Mirror

Real-time video attention tracking tool utilizing browser-side computer vision inference for live audience engagement measurement.

Blood Bank Demand-Forecasting System

URL: /projects/blood-bank/ | .md Mirror

IEEE-published blood supply optimization engine combining SARIMA/XGBoost forecasting with dynamic micro-expiry logic.

A Demand-Driven Software Approach with Dynamic Micro-Expiry and Just-in-Time Processing to Reduce Platelet Wastage in Blood Banks

Venue: IEEE | Link: IEEE Xplore #11584332

Authorship & Contribution: Sahil Tomar (Co-Author) — Responsible for ideation, problem formulation, SARIMA/XGBoost model development, and simulation software execution.

• Reduced simulated platelet wastage from 11.2% to 2.5% (~78% relative reduction).
• Maintained 99.1% demand fulfillment rate.
• SARIMA model achieved lowest MAE (5.85) among tested models.
• Results validated across 30 iterations via paired t-tests (p < 1.22×10⁻¹²).