AI SYSTEMS ENGINEER // SHIVAPRASAD R MDavanagere, IN

Architecting Production AI Systems & LLM Infrastructure.

I build production SaaS, multi-tenant AI platforms, shadow eval engines, and secure RAG knowledge stores — shipped independently with zero shortcuts.

Production LLM SystemsAutonomous AI AgentsShadow Model EvalsRAG & Vector DBsNext.js 16 / SaaSHMAC Security Loops
7+
Production Systems
68%
LLM Cost Savings
99.4%
OCR Accuracy
<5ms
Proxy Latency
LIVE SYSTEM TELEMETRY
OPERATIONAL
>$ tokenlens proxy --mode shadow-eval
40-70%
Cost Reduction
100%
Source Grounded
HMAC-SHA256
Security Gate
自己紹介 // ENGINEERING DOSSIER

Building Production AI Systems from First Principles

I'm an AI systems engineer based in Davanagere, Karnataka— a Tier 2 city in India. I mention this because it defines my trajectory: I've architected and shipped complete production systems independently, without a venture-backed office or large dev team.

My workflow is AI-native from the ground up. I leverage Claude Code, Cursor, and Antigravity as a core engineering methodology. Solo, I shipped three live multi-tenant SaaS platforms, built a live LLM optimization proxy (TokenLens), engineered a Telegram-native procurement agent with an auditable learning loop (Uneora), and authored ai-map on npm.

Currently, I'm building TokenLens and spec'ing Shruta (Company Brain, aligned with YC S26 RFS). I'm applying to YC W27 and exploring Founding Engineer, AI Systems, or Lead Product roles where I can operate at maximum velocity.

Location Davanagere, IN
Development Mode AI-Native Solo
Target Horizon YC W27 Batch

LLM Infrastructure & Evaluation

Production proxy systems, automated shadow model evaluations, cost optimization proxies, and prompt compression at scale.

Autonomous AI Agent Systems

Multi-tier approval agents, learning loops with noisy-data guardrails, HMAC-secured Telegram bots, and versioned per-org parameters.

Knowledge & Vector RAG Architecture

PGVector database design, Company Brain engines, contextual document overlap chunkers, and executable skills specs.

Production SaaS Engineering

End-to-end full-stack delivery: Next.js 16, PostgreSQL RLS, Cloudflare Workers, Upstash Redis, Razorpay, and multi-tenant security.

プロダクト // FEATURED SYSTEMS SHOWCASE

Production Systems & AI Architecture

LIVE SaaSL1
Live App

TokenLens

[T1_TOKENLENS]

AI Cost & ROI Optimization proxy with shadow model evaluation

LIVE COST & EVAL SIMULATOR
PROPRIETARY PROXY
Daily API Volume:500,000 tokens/day
Shadow Target Model:
Projected Monthly Savings$56 / mo
Cost Reduction75% DOWN
COST_REDUCTION40-70%
SETUP_TIME<5 min
Inspect Architecture Deep Dive
LIVE SaaSL1
Live App

Uneora v2

[T1_UNEORA_V2]

Multi-tenant unified commerce platform for MSMEs with AI agent learning loop

AUTONOMOUS AGENT LOOP
HMAC SECURED
OCR INGEST
PDF Bill -> extracted 99.4%
HMAC CHALLENGE
Generated 48h token
TG CALLBACK
Awaiting human tap
PARAM UPDATE
Learning loop committed
Noisy-Data Guardrail StatusACTIVE — 100% AUDITABLE RLS PARAMS
DAILY_ORDERS342
STORES_SYNCED23
Inspect Architecture Deep Dive
SPEC COMPLETEL1

Shruta (Company Brain)

[T1_SHRUTA]

Secure, RAG-based institutional knowledge platform grounding AI agents

  • Ingests and normalizes fragmented emails, Slack message threads, and PDF SOPs.
  • Embeds documents into pgvector/Pinecone with custom metadata filters.
  • Retrieval pipeline optimization featuring chunking strategies, reranking, and context grounding.
YC_ALIGNMENTS26 RFS
RETRIEVAL_LATENCY<200ms
Inspect Architecture Deep Dive
LIVE SaaSL2
Live App

Vyntra HRMS

[T2_VYNTRA]

Workforce and field-force HRMS with real-time GPS tracking

  • Real-time GPS coordinates upload, route deviation alerts, and historical replay.
  • Geofenced check-in zones with required photo capture proof.
  • Automated payroll calculating PF, ESI, TDS, and LOP deductions in one click.
FIELD_RECORDS24k+
PAYROLL_SAVINGS20h/mo
Inspect Architecture Deep Dive
LIVEL2
Live App

TEZride

[T2_TEZRIDE]

Two-sided ambulance booking marketplace in India

  • SOS emergency booking flows and operator booking dispatch dashboards.
  • Supports BLS, ALS, ICU, and Neonatal specialized ambulance types.
  • Real-time tracking of dispatched ambulances and patient coordinates.
DISPATCH_TIME<8 min
ACTIVE_FLEET64
Inspect Architecture Deep Dive
SEO READYL2
Live App

Sohman Analytical

[T2_SOHMAN]

High-performance SEO architecture and analytics dashboard built for clients

  • Static site generation with incremental regeneration (ISR) delivering <500ms load times.
  • Rich JSON-LD schemas, sitemaps, and optimized semantic structure targeting rapid indexation.
  • Integrated SEO indexation monitoring dashboard checking search console status.
LIGHTHOUSE100/100
INDEXATION<24 hours
Inspect Architecture Deep Dive
PUBLISHEDL3
Live App

ai-map

[T3_AI_MAP]

Open-source context window and token optimization package

  • Context compression and memory continuity algorithms.
  • Reduces token expenditures in agentic code loops by 35%.
DOWNLOADSWeekly
Inspect Architecture Deep Dive
ACTIVE DEVL3

ai-agents

[T3_AI_AGENTS]

HMAC-secured Telegram agent manager with learning loops

  • HMAC-SHA256 signature verification.
  • Learning loop with noisy-data filters.
SECURITYHMAC-SHA
Inspect Architecture Deep Dive
WAITLISTL3
Live App

Candid

[T3_CANDID]

Premium, pleasure-positive adult intimacy dating landing page

  • High-speed landing page with clean accessibility markup.
  • Waitlist verification and automations using Netlify Functions.
WAITLIST_SIZE850+
Inspect Architecture Deep Dive
冒険 // Side Adventures

Technical Side Adventures

Targeted technical assignments, research projects, and specialized AI/ML pipelines built under contract constraints.

AI Video Dubbing Pipeline

Contract AI/ML Engineer
[SATVIC_DUB]

Built as a production prototype for Satvic Foods to automate video translation into South Indian languages with colloquial dialect matching.

  • Implemented speaker diarization with word-level timestamps to detect speakers and timing parameters.
  • Engineered a duration-aware rewrite loop using LLMs to translate to colloquial Kannada/Telugu/Malayalam within +/-15% timing tolerance.
  • Designed automated QC flow with back-translation semantic comparisons and timing drift measurement.
  • Integrated glossary locks to ensure brand terms remain consistent across translated audio segments.
PythonWhisper ASRpyannote.audioSarvam TTSElevenLabsFFmpeg

SLM Fine-Tuning Pipeline

Contract ML Engineer
[BAYARD_FINE]

Delivered as a rigorous ML Engineering assignment for Bayard Vacations, showcasing end-to-end model tuning, evaluation, and test coverage.

  • Fine-tuned DistilGPT2 on 15,011-row Databricks Dolly-15k using LoRA rank 16, alpha 32.
  • Created CLI inference client supporting temperature, seeding, and token generation controls.
  • Developed 10-test pytest suite validating determinism, weight leakage, boundary overflow, and overfitting.
  • Conducted qualitative evaluation of response characteristics under varying hyperparameters.
PythonHuggingFacePEFT (LoRA)PyTorchpytestDistilGPT2
経験 // Experience

Experience History

Typographic chronicle of start-up operations, full-stack systems engineering, and procurement analytics.

012024 — Present[SWEDANA_DEV]

Founder & AI Systems Engineer

Swedana — Independent Product Studio

  • Independently designed, built, and deployed three production SaaS platforms: VYNTRA, Uneora, and TEZride, owning full-stack development, PostgreSQL RLS, and infrastructure.
  • Built TokenLens — a live AI spend optimization proxy platform featuring automated quality scoring and background shadow evaluations.
  • Architected Uneora's AI procurement agent with Telegram approvals, HMAC-secured callbacks, noisy-data filtering, and versioned parameters.
  • Published ai-map on npm — an open-source context management utility reducing agent token waste by 35% in long sessions.
02Aug 2025 — Feb 2026[LC_OPS]

Business Research & Product Operations Analyst

Living Canvas LLP — AI-first early-stage startup

  • Owned entire product flow for IIN (IndiaMART-style B2B directory) — user flows, wireframes, navigation architecture, feature development, and end-to-end UAT
  • Wireframed and UAT-tested an industrial IoT emissions monitoring app (O2, CO2, particulates); documented 50+ bugs across 3 platform products
  • Designed AI-assisted intake and classification pipelines for a McMaster-Carr India equivalent — minimizing human review with human-in-the-loop approval gates
  • Operated inside an AI-first culture — used Claude, Cursor, and Antigravity across every workstream to accelerate research, documentation, wireframing, and QA
03Feb 2025 — Aug 2025[IBID_OPS]

Operations & Sales Lead

Ibid Autoparts

  • Managed end-to-end operations and sales for a US-focused auto parts business
  • Handled client communication, onboarding, pricing negotiations, and delivery tracking across time zones
  • Coordinated with US-based partner on lead management and marketing strategy
04Jan 2023 — Jan 2025[VM_PROC]

Procurement Manager

Venkatesh Medico Pvt. Ltd.

  • Built Tableau dashboards to visualize procurement trends — reduced excess inventory by 25% in 6 months
  • Automated reporting using Tableau and Advanced Excel, saving 20+ hours monthly
  • Analyzed procurement data to optimize inventory levels, reducing overhead costs by 10%
  • Designed forecasting dashboards to align inventory planning with business objectives
技術 // ARCHITECTURAL SKILL MATRIX

Technical Matrix & Tools

LLM Infrastructure & Evals

Claude 3.7 APIOpenAI APIShadow Model EvalsToken Optimization ProxyPrompt Compressionpgvector Indexingai-map OSS

Full-Stack SaaS & Cloud

Next.js 16 (App Router)TypeScriptPostgreSQL (Neon RLS)Prisma ORMCloudflare WorkersUpstash RedisRazorpay / PaymentsNextAuth v5

AI-Native Builder Stack

Claude CodeAntigravity IDECursorAutonomous Agent LoopsRAG Vector ArchitectureHMAC Callback Gates

Data Systems & Operations

SQL AnalyticsTableau DataVizAdvanced Financial ModelingVendor ProcurementProcess Engineering
連絡 // Contact

Get in Touch

Open to founding engineer, AI systems, and senior PM roles. Based in Davanagere, Karnataka — available remotely worldwide.

Curriculum Vitae

Resume / CV

My complete professional history, technical stack, and engineering accomplishments.