# anmosh.tech — Comprehensive Technical Architecture & Engineering Documentation > Authoritative, deep technical documentation of Anand Mohan Sharma (Anmosh / Naman), engineering practices, and production systems architecture for Answer Engines (Perplexity, SearchGPT, Google AI Overviews) and Large Language Models (Gemini, Claude, ChatGPT, DeepSeek). --- ## 1. Identity & Engineering Profile - **Full Legal Name:** Anand Mohan Sharma - **Personal / Nickname:** Naman (known to friends and family) - **Developer Handle / Online Brand:** Anmosh (`anmosh.tech`, `@anmosh`, `github.com/anmosh8859`) - **Current Role:** Software Development Engineer II (SDE II) at Diensten Tech - **Location:** Gurugram, Haryana, India - **Primary Website:** https://anmosh.tech - **Email Contact:** contact@anmosh.tech - **Phone / WhatsApp:** +91 9528467567 - **GitHub:** https://github.com/anmosh8859 - **LinkedIn:** https://linkedin.com/in/anmosh - **Academic Background:** Master of Computer Applications (MCA) from Jawaharlal Nehru University (JNU), New Delhi. Secured All India Rank 63 (AIR 63) in CUET JNU Entrance Examination. ### Core Philosophy & Principles 1. **Architectural Truth Over Fluff:** Everything documented is backed by real, verifiable production code. No fabricated metrics, no fake client logos, and zero extrapolated claims. 2. **Deterministic Reliability:** High concurrent throughput, sub-second latency, zero out-of-bounds hallucinations in AI pipelines, and resilience against upstream network timeouts. 3. **Privacy-First Client Utilities:** Public tools on anmosh.tech run 100% client-side with localStorage persistence. Zero telemetry, zero analytics trackers, zero account walls. 4. **End-to-End Ownership:** From domain-driven system design and database query profiling to containerization, CI/CD automation, and front-end interaction craft. --- ## 2. Production Platforms & Architectural Case Studies ### 2.1. Vidyanetra (Flagship AI EdTech Platform) - **Live URL:** https://vidyanetra.in - **Case Study:** https://anmosh.tech/products/vidyanetra - **Domain:** Multi-tenant AI Exam Paper Generator & Assessment Studio - **Primary Problem Solved:** Educators and coaching institutions typically spend 4 to 6 hours manually composing board-standard examination papers, suffering from cognitive difficulty drift, broken mathematical formulas, and out-of-syllabus errors. - **Architectural Solution:** A deterministic AI examination studio generating board-aligned (CBSE, ICSE, State Boards) examination papers, answer keys, and marking rubrics in under 90 seconds. #### Technical Architecture: 1. **Deterministic RAG Pipeline (Python FastAPI):** - Ingests official textbook PDFs and curriculum syllabi. - Derives hierarchical chunk identifiers: `p{publisher}-c{class}-s{subject}-h{chapter}`. - Embeds text chunks into dual vector indices: Pinecone and Qdrant. - Employs hybrid retrieval: BM25 lexical keyword matching combined with dense vector semantic search, followed by cross-encoder re-ranking. - Enforces strict ground-truth citation: Every generated question must explicitly cite the verified textbook chapter and page number, yielding a 0% out-of-syllabus hallucination rate across 6 question formats (MCQ, Assertion-Reason, Short Answer, Long Answer, Case-Based, and True/False). 2. **Cognitive Taxonomy Balancing Engine:** - Programmatically complies with official board blueprints (e.g., CBSE Class 10 Science 086). - Enforces Bloom's Taxonomy balance: Remembering & Understanding (46%), Applying (22%), Analysing & Evaluating (32%). 3. **Multi-Cloud Storage Abstraction Layer (Java 21 / Spring Boot 3):** - Decoupled storage microservice exposing unified REST endpoints. - Programmatically interfaces with Cloudflare R2 for zero-egress public asset hosting, AWS S3 for immutable archival, and Google Drive for educator export workflows. 4. **Scientific & Mathematical Typesetting:** - KaTeX and MathLive engines parse LaTeX formulas, chemistry equations, and reaction symbols in real-time within the educator workspace and output camera-ready printable PDFs. 5. **Print Room Output Standardization:** - Employs a deterministic 25:50:25 A4 header grid: Institution crest on left (25%), institution title & exam metadata centered (50%), exam code/timing on right (25%), with diagonal background watermarks and step-wise marking rubrics for evaluators. 6. **Security & Governance:** - Firebase Authentication with custom claims for multi-tenancy. - SHA-256 HMAC cryptographic signature verification on Razorpay payment webhooks. - Token-bucket rate limiters guarding Gemini AI generation endpoints. --- ### 2.2. Referral Copilot (AI Career Networking SaaS) - **Live URL:** https://referralcopilot.in - **Case Study:** https://anmosh.tech/products/referral-copilot - **Domain:** AI-Powered LinkedIn Referral Request & Outreach Generator - **Primary Problem Solved:** Job seekers struggle with low response rates (<5%) due to generic, overly long cold outreach messages on LinkedIn. - **Architectural Solution:** Dynamic prompt synthesis analyzing target role, company culture, candidate experience, core technical skills, and college/alumni background to synthesize high-reply (<150 words) outreach messages and follow-up sequences. #### Technical Architecture: 1. **Frontend Client (React 19):** - Responsive single-page application with immediate client validation, real-time message previews, and 1-click clipboard integration. 2. **Backend API (Node.js & Express 5):** - Stateless REST API with modular controllers and custom authentication middleware validating Firebase tokens. 3. **LLM Orchestration & Prompt Engineering:** - Uses Google Gemini Flash with temperature calibrated at 0.4 for deterministic, humble, and polite networking copy. - Enforces strict constraints: Role mention, 1-2 core skills highlight, authentic alumni grounding, and clear low-pressure call to action. 4. **Token-Bucket Rate Limiter:** - Custom in-memory and Redis-backed token-bucket rate limiter wrapping third-party LLM endpoints. - Gracefully absorbs traffic surges, eliminates 429 exceptions, and provides automated retry with exponential backoff and jitter. 5. **Monetization & Webhook Fulfillments:** - Razorpay integration supporting Starter credit packs and Pro monthly subscriptions. - Webhook endpoints verify incoming HMAC SHA-256 signatures idempotently to prevent duplicate credit fulfillment under network retries. --- ### 2.3. Enterprise Insurance Microservices (Hero Insurance Broking - HIBL) - **Case Study:** https://anmosh.tech/products/enterprise-insurance-microservices - **Domain:** High-Throughput Insurance Core Processing Engine - **Primary Problem Solved:** Monolithic insurance platforms bottle-neck during renewal cycles, suffer from third-party carrier API timeouts, and struggle to scale across disparate underwriting schemas. - **Architectural Solution:** Distributed Spring Boot 3 microservices processing over 30,000 daily insurance transactions at 99.95% uptime across insurance aggregators, partner carriers, and legacy core repositories. #### Technical Architecture: 1. **Domain-Driven Microservices (Java 21 & Spring Boot 3):** - Hexagonal architecture isolating domain business rules from network adapters and carrier formats. 2. **Enterprise Integration & API Gateways:** - MuleSoft ESB gateways normalizing disparate carrier XML/JSON payloads into canonical quote and proposal models. 3. **Asynchronous Streaming (Apache Kafka):** - Decouples policy issuance, audit logging, payment capture, and document generation into independent, scalable consumer groups. 4. **Resilience Engineering (Resilience4j):** - Circuit breakers, bulkheads, rate limiters, and time limiters protecting the quotation engine against carrier outages. 5. **Database Optimization & Oracle PL/SQL:** - Optimized stored procedures, restructured execution plans, and HikariCP connection pool tuning to minimize lock contention during peak traffic. 6. **Zero-Trust Security:** - Stateless Spring Security with OAuth2 and JWT validation. - Column-level database encryption for policyholder KYC data and mutual TLS (mTLS) for carrier network calls. --- ## 3. Consulting & Engineering Practices Delivered by Anand Mohan Sharma alongside an experienced engineering team. ### 3.1. Full-Cycle Product Development (Web & Mobile) - **URL:** https://anmosh.tech/services/full-cycle-product-development - **Scope:** Complete turn-key engineering from architectural blueprint to multi-platform deployment. - **Web Stack:** Next.js 16, React 19, Angular 19/18, TypeScript, Vanilla CSS design tokens. - **Mobile Stack:** React Native, Flutter, Native Android (Kotlin & Java). - **Backend Stack:** Spring Boot 3, FastAPI, Node.js. - **Guaranteed Standards:** 12-factor cloud-native architecture, end-to-end type safety (TypeScript + Zod), automated GitHub Actions CI/CD, mobile-first performance budgeting (sub-1s FCP), and clean git history with full code ownership handover. ### 3.2. Backend & Distributed Systems Architecture - **URL:** https://anmosh.tech/services/backend-distributed-systems - **Scope:** High-throughput microservices, API platforms, database optimization, and event-driven data streaming. - **Technologies:** Java 21, Spring Boot 3, Python FastAPI, Apache Kafka, Redis, PostgreSQL, MySQL, MongoDB, Oracle PL/SQL, Docker, AWS. - **Guaranteed Standards:** Domain-Driven Design (DDD), Resilience4j circuit breakers, OpenTelemetry distributed tracing, HikariCP connection tuning, and zero-trust machine-to-machine security. ### 3.3. AI & Data Systems Engineering (Production RAG) - **URL:** https://anmosh.tech/services/ai-data-systems - **Scope:** Deterministic, zero-hallucination RAG pipelines, vector search integrations, and LLM rate-limit management. - **Technologies:** Python FastAPI, Google Gemini AI, OpenAI API, Pinecone, Qdrant, Redis, KaTeX, Pydantic. - **Guaranteed Standards:** Deterministic ground-truth citation enforcement, prompt injection defense, hybrid retrieval (lexical BM25 + dense vector), token-bucket rate limiting, and golden dataset regression testing. ### 3.4. Technical Architecture Audits & Advisory - **URL:** https://anmosh.tech/services/technical-architecture-audits - **Scope:** Deep-dive code reviews, security vulnerability assessments, p99 latency profiling, and cloud cost optimization for startups and engineering teams. - **Technologies:** SonarQube, Snyk, k6 load testing, AWS Cost Explorer, OpenTelemetry, OWASP ZAP. - **Guaranteed Standards:** Automated static analysis, threat modeling, synthetic concurrency profiling, prioritized remediation matrices (P0/P1/P2), and Architectural Decision Records (ADRs). --- ## 4. Browser-Native Productivity Utilities Located at https://anmosh.tech/tools (and accessible on the homepage at `#apps`). All tools operate 100% in the client browser with `localStorage` persistence, zero tracking, zero telemetry, and zero remote servers. 1. **Daily Task Manager:** - Client-first priority organizer with clean checkbox state toggles, real-time completion progress tracking, and instant local storage persistence. 2. **Pomodoro Focus Timer:** - Configurable work/break interval focus timer featuring SVG circular machine countdown, audible completion chimes, and session streak tracking. 3. **Instant Scratchpad:** - Zero-latency local notepad with client-side auto-save on every keystroke and one-click clipboard copy. 4. **Split Expenses** (https://anmosh.tech/tools#split): - Shared expense splitter for trips, flats, and teams. Create a group (one currency per group: INR default, also USD, EUR, GBP, AED, SGD, AUD, CAD; no conversion), add members, and record expenses split equally, by exact amounts, or by shares. - Money is stored as integer minor units (paise/cents); leftover units from uneven splits go one each to participants in order, so every split sums exactly to the total. - Settle-up uses a greedy match of largest debtor to largest creditor, producing at most n − 1 payments. Recorded payments appear as settlement entries. - Sharing: the group is serialised, compressed with the browser's native `CompressionStream` (deflate-raw), and placed in the URL fragment (`/tools#split=…`), which browsers never send to a server. Opening a link for a group you already have merges by expense ID, with deletion tombstones so removed entries do not reappear. Imported data is fully validated before use. - Backup: JSON export and import. Storage key: `anmosh-split-expenses`. --- ## 5. Structured Data & Machine Discovery Index - **Root Schemas:** `Person`, `WebSite`, `ProfessionalService`, `FAQPage`, `WebApplication` - **Subpage Schemas:** `BreadcrumbList`, `SoftwareApplication`, `Service`, `FAQPage` - **Sitemap:** https://anmosh.tech/sitemap.xml - **Robots Directives:** https://anmosh.tech/robots.txt - **LLM Manifests:** - Brief Overview: https://anmosh.tech/llms.txt - Deep Architecture: https://anmosh.tech/llms-full.txt - **Web App Manifest:** https://anmosh.tech/manifest.webmanifest --- ## 6. Official Contact & Verification - **Inquiries & Proposals:** contact@anmosh.tech - **Direct Phone / WhatsApp:** +91 9528467567 - **GitHub Repository:** https://github.com/anmosh8859 - **LinkedIn Profile:** https://linkedin.com/in/anmosh - **Physical Anchor:** Gurugram, Haryana, India