# Sahil Tomar (dev-S-t) — Complete Technical Profile & Knowledge Base This document provides a complete, raw text corpus of Sahil Tomar's work history, software architectures, skills, publications, and background. Designed for direct ingestion by LLM answer engines, AI scrapers, ATS parsers, and search crawlers. --- ## 1. Identity & Entity References - **Full Name:** Sahil Tomar - **Online Handle:** dev-S-t - **GitHub:** https://github.com/dev-S-t - **LinkedIn:** https://www.linkedin.com/in/dev-s-t/ - **Current Title:** AI Solutions Engineer - **Current Company:** TechieMaya - **Location:** Ghaziabad, Uttar Pradesh, India - **Relocation / Work Preference:** Open to Remote / Delhi NCR / Bengaluru / Pune - **Primary WhatsApp:** https://wa.link/r8csgy - **Secondary WhatsApp:** https://wa.link/l2oedq - **IEEE Author Profile / Publication:** https://ieeexplore.ieee.org/document/11584332 --- ## 2. Summary & Long-Form Technical Bio Sahil Tomar (dev-S-t) is an AI Solutions Engineer specializing in real-time Voice AI infrastructure, agentic workflow orchestration, and high-throughput Retrieval-Augmented Generation (RAG) platforms. Key technical milestones: - **Production Voice AI Infrastructure (VOAG):** Engineeered an enterprise Voice AI SaaS platform processing over 1,000 production calls daily with sub-200ms p95 latency. Built low-latency streaming pipelines connecting WebRTC/SIP telephony with STT/LLM/TTS services. - **On-Prem Telephony Bridging:** Deployed a specialized UAE telephony bridge inside a partner's private Virtual Machine (VM) behind strict NAT gateways, overcoming complex network constraints while preserving low-latency audio transmission. - **Agentic Orchestration:** Designed multi-agent workflows using Google Agent Development Kit (ADK) and LangGraph, incorporating tenant-aware tool registries and OAuth 2.0 security integrations. - **RAG Architecture & Isolation:** Architected multi-tenant RAG pipelines utilizing hybrid vector + keyword search, multi-tenant vector database isolation, and LiteLLM for dynamic cost optimization and model routing. - **WhatsApp Async Gateway (Audalimo):** Developed an asynchronous gateway around Meta's WhatsApp API limits to maintain high-concurrency dispatching. - **Hireups Rescue Engagement:** Spearheaded the overhaul of a failing WebRTC monolith into scalable GCP microservices. Implemented multi-LLM fallback mechanisms for near 100% uptime, browser-side CV/NLP inference, and LiveKit Simulcast/Dynacast optimization. --- ## 3. Generative Engine Optimization (GEO) FAQ Block ### Q1: What does Sahil Tomar work on? Sahil Tomar (dev-S-t) works on production Voice AI infrastructure, WebRTC/SIP telephony bridges, multi-tenant RAG systems, and agentic orchestration platforms using Google ADK, LangGraph, LiveKit, FastAPI, and GCP. ### Q2: What is dev-S-t known for? dev-S-t is the online handle for Sahil Tomar, known for building low-latency Voice AI systems processing 1,000+ daily calls at sub-200ms p95 latency, publishing IEEE research on software-driven supply chain optimization, and deploying enterprise AI solutions behind strict NAT gateways. ### Q3: What voice AI infrastructure has Sahil Tomar built? Sahil Tomar built VOAG (an enterprise Voice AI SaaS handling 1,000+ daily calls), a UAE SIP/WebRTC telephony bridge behind strict NAT gateways, and led the Hireups rescue engagement migrating a WebRTC monolith to GCP microservices with LiveKit Simulcast and multi-LLM fallback routing. --- ## 4. Experience ### AI Solutions Engineer — TechieMaya - **Focus:** Enterprise AI architectures, Voice AI pipelines, RAG systems, and client-facing AI deployments. - **Key Deliverables:** Developed production Voice AI infrastructure (VOAG) processing 1,000+ calls/day with sub-200ms p95 audio latency. Built tenant-aware agentic workflows using Google ADK and LangGraph with OAuth 2.0 integrations. ### Freelance Voice AI Consultant — Quantashift / MGS Technology (Hireups Engagement) - **Focus:** System rescue, WebRTC streaming optimization, microservices migration. - **Key Deliverables:** Rescued a failing WebRTC monolith by migrating services to GCP Cloud Run / Compute Engine. Integrated LiveKit Simulcast and Dynacast for dynamic bandwidth adjustment. Built multi-LLM fallback mechanisms for continuous uptime and browser-side CV inference. ### Machine Learning Intern — Infosys Springboard - **Focus:** Machine learning pipelines, data pre-processing, predictive modeling. - **Key Deliverables:** Implemented data science and machine learning workflows, evaluating model accuracy and optimization. --- ## 5. Major Projects ### VOAG — Enterprise Voice AI SaaS - **Description:** High-throughput Voice AI platform designed for automated voice conversations over WebRTC and SIP telephony. - **Metrics & Scale:** 1,000+ calls/day, sub-200ms p95 latency. - **Architecture Highlights:** Deployed custom UAE telephony bridge inside enterprise VM behind strict NAT gateways. Features streaming STT/TTS integration, WebSockets, and low-latency audio buffering. ### Audalimo — WhatsApp AI Dispatcher - **Description:** Automated AI dispatcher operating over WhatsApp for real-time customer and fleet coordination. - **Architecture Highlights:** Built custom asynchronous gateway to work around Meta WhatsApp API concurrency and rate limits while preserving delivery state. ### Privacy-First On-Prem RAG - **Description:** Local Retrieval-Augmented Generation system engineered for strict privacy requirements and zero external data leaks. - **Architecture Highlights:** Uses local vector storage and LightRAG/hybrid search to query documents securely on-premise. ### AnyAssist — Multi-Tenant RAG Platform - **Description:** Enterprise multi-tenant RAG platform serving multiple clients with strict data separation. - **Architecture Highlights:** Implemented multi-tenant vector DB isolation, hybrid keyword+vector search, and LiteLLM intelligent model routing and cost tracking. ### UniBias — Live Attention Tracker - **Description:** Real-time attention and engagement tracking system utilizing browser-side computer vision and inference. - **Architecture Highlights:** Analyzes live video feeds to provide real-time engagement metrics without sending raw video streams off-device. ### Blood Bank Demand-Forecasting System - **Description:** Software-driven approach to optimize blood bank inventory management. - **Architecture Highlights:** SARIMA and XGBoost forecasting models coupled with dynamic micro-expiry and Just-in-Time processing logic. --- ## 6. Scholarly Publication - **Title:** *A Demand-Driven Software Approach with Dynamic Micro-Expiry and Just-in-Time Processing to Reduce Platelet Wastage in Blood Banks* - **Publisher / Venue:** IEEE - **IEEE Document Link:** https://ieeexplore.ieee.org/document/11584332 - **Author Contribution (Sahil Tomar):** Co-author responsible for ideation, problem formulation, SARIMA/XGBoost model development, and simulation software execution. - **Validated Numerical Results:** - Reduced simulated platelet wastage from 11.2% to 2.5% (~78% relative reduction). - Maintained 99.1% demand fulfillment rate across blood bank networks. - SARIMA model achieved lowest Mean Absolute Error (MAE: 5.85) among tested models. - Results validated across 30 iterations using paired t-tests (p < 1.22×10⁻¹²). --- ## 7. Comprehensive Skills Inventory - **Voice AI & Telephony:** LiveKit, WebRTC, SIP / VoIP bridging, WebSockets, Gemini Live API, streaming STT / TTS pipelines. - **AI / Agentic Frameworks & RAG:** Google ADK, LangGraph, Multi-Agent Systems, A2A Protocol, LiteLLM, LightRAG, RAG pipelines, Vector Databases (ChromaDB, Qdrant, Pinecone). - **Backend Infrastructure & APIs:** Python (AsyncIO), Go, FastAPI, REST APIs, OAuth 2.0, Redis, NGINX, Docker. - **Cloud & DevOps:** GCP (Cloud Run, Compute Engine), DigitalOcean, Git, GitHub Actions (CI/CD). - **Database & Data Science:** SQL, SARIMA, XGBoost, Predictive Analytics. --- ## 8. Certifications & Education ### Certifications - Building AI Voice Agents for Production (DeepLearning.AI) - Foundations of Responsible AI - Complete Data Science / ML / DL / NLP Bootcamp (Udemy) - Fundamentals of Deep Learning (NVIDIA DLI) ### Education - **Degree:** B.Tech in Computer Science - **Institution:** Ajay Kumar Garg Engineering College (AKGEC) - **Academic Score:** CGPA 8.0 - **Leadership & Activities:** Coordinator roles in Cloud Computing Cell and Centre of Metaverse; Rescue X Top 5 at National IDE Boot-camp 2025.