Siddharth

Hi, I am Sidharth Rodrigues

Senior AI Engineer

Mumbai · Remote

Lifted team PR throughput 2.5x at Airawath (93 to 222 merged PRs in ~11 weeks) by building an internal Claude Code agent suite — 7-specialist code review + adversarial test writers. 4+ years shipping production agentic systems, hybrid RAG pipelines, and SOC2-grade infra; $169K+ in cumulative client savings. Currently runs day-to-day delivery (reviews, deploys, planning, hiring) at an AI startup.

Experience

Roles and projects that shaped my journey as a developer.

AI Tech Lead

Airawath
Feb 2026 - Present
Built internal Claude Code agent suite (7-specialist code review + adversarial test writers); team PR throughput up 2.5x (93 to 222 merged PRs in ~11 weeks). Shipped Talk to Elie AI assistant, Notification platform, CMS. Owns SOC2/ISO27001 audit prep, AWS modernization, and day-to-day delivery (reviews, deploys, hiring)
Claude CodeClaude (Opus 4.7 / Sonnet 4.6 / Haiku 4.5)GitHub Actions OIDCGHCRAWS (Secrets Manager, CloudFront, Route 53, IAM, GuardDuty, CloudTrail)LinearMulti-agent orchestrationAWS SES / Twilio / Firebase

Founding Engineer

CrazyTok Media
Aug 2025 - Feb 2026
Engineered agentic video QC pipeline (custom MobileNetV3 caption prefilter saving 93% API cost, EfficientNet-B0/K-means B-roll detector, 3-tier Gemini 2.5 Pro fallback) cutting 60min to 5min per video at $0.48/video (~50x cheaper than $15-30 manual, $10K+/yr saved). Built 5-stage agentic content generation pipeline; drove Claude Code adoption across founder and ops team
Founding Engineer architecture diagram
Click to enlarge
PythonGPT-4 & Gemini APIsComputer Vision (OpenCV)LangChainChrome Extension DevelopmentAudio AnalysisPrompt EngineeringMachine Learning

AI Automation Engineer

RSL Media Hub
Jan 2025 - Dec 2025
Built end-to-end agentic blog pipeline (9-stage Gemini workflow: keyword discovery, SERP/PAA/competitor brief, drafting with internal linking, multi-image generation, Sanity CMS publish) at $0.35/blog with daily GCP cron + GHA failover. Implemented Voice AI on GoHighLevel (95%+ intent accuracy, <2s latency); built LinkedIn lead enrichment pipeline processing 14K+ businesses via async Python + Crawl4AI + Gemini
AI Automation Engineer architecture diagram
Click to enlarge
Voice AI & Speech RecognitionGoHighLevel CRMPython 3.8+ with asyncioCrawl4AI (web scraping)Google Gemini 2.5 FlashChatGPT APIMake.com (workflow automation)Next.js / Modern Web Frameworks

Software Integration Engineer

Factories of Future (TVS Motor)
Feb 2025 - Aug 2025
Built real-time MQTT-to-REST protocol bridge integrating AMRs into Twinzo digital twin: 10Hz position streaming, 50-80ms latency, OAuth 2.0 per-device auth (99.9% success), +/-20mm coordinate transform vs +/-50mm spec. Coordinated across 4 organizations (TVS Motor, Hi-tech Robotics, Twinzo, FoF); authored 50+ pages of technical documentation
Software Integration Engineer architecture diagram
Click to enlarge
PythonMQTT (Mosquitto/HiveMQ)REST APIOAuth 2.0DockerPostgreSQLRedisReal-time Data Streaming

Software Engineer

Viven Eduversity (Govt. affiliated client)
May 2024 - Mar 2025
Led small dev team building distributed AWS web scraping system: 4-month manual collection (80 people) to 2-week automated pipeline (87.5% reduction, $136K annual savings). 50x performance via Selenium Hub + multi-threading (4,000 records/hour). Tesseract + EasyOCR image processing modules at 97% accuracy
Software Engineer architecture diagram
Click to enlarge
AWS (EC2, Lambda, S3, CloudWatch)PythonSelenium WebDriverBeautifulSoupPandasMachine Learning (CAPTCHA solver)DockerPostgreSQL

Projects

Maritime Dark Ship Detection - Multi-Sensor Fusion & RAG System

Real-time multi-sensor fusion detecting AIS-evading vessels with hybrid RAG pipeline for maritime intelligence queries
Maritime Dark Ship Detection - Multi-Sensor Fusion & RAG System architecture diagram
Click to enlarge
Python 3.11+FastAPINext.js 14Three.jsPostgreSQL + pgvectorRedis StreamsGoogle Gemini 2.5 (Flash & Pro)LangChainSciPy (Hungarian algorithm)PydanticWebSocket

Alive AI - Brain-Inspired 7-Loop Architecture for Real-Time Embodied AI

Brain-inspired 7-loop architecture (CfC continuous-time backbone + LoRA-wrapped SmolLM2-360M) with dynamic compute allocation, state-crystallization gradient (VICReg, Mamba-style input-dependent gating, predictive-coding multi-timescale loss), and 7 ordered falsification gates. LoRA backbone validated at 0.999x baseline perplexity
PyTorchCfC (continuous-time)LoRA / PEFTSmolLM2-360MVICRegMamba-style gatingPredictive codingRTX 4080

PageResUNet - Deep Learning Model for OCR Preprocessing & Document Analysis

Research and development of PageResUNet, a deep learning model for optical character recognition and document layout analysis, combining ResNet and U-Net architectures for superior accuracy.
PageResUNet - Deep Learning Model for OCR Preprocessing & Document Analysis architecture diagram
Click to enlarge
PyTorchDeep LearningResNetU-NetComputer VisionOCRPython

LinkedIn Lead Enrichment Pipeline

Enterprise lead generation, 14,260 businesses, 99.98% completion, 100 concurrent tasks
LinkedIn Lead Enrichment Pipeline architecture diagram
Click to enlarge
Python 3.8+Crawl4AI v0.7.4Google Gemini 2.5 Flashasyncio (concurrent processing)JSON/CSV/Markdown outputWeb Scraping at Scale

Agentic Blog Pipeline - 9-Stage Gemini Workflow

End-to-end agentic blog pipeline: 9-stage Gemini workflow (keyword discovery, SERP/PAA/competitor brief, drafting, multi-image generation, Sanity CMS publish), funnel-aware routing (TOFU/MOFU/BOFU), 452-line brand-voice prompt, retry + state-recovery. ~$0.35/blog at 20-30min runtime; daily cron on GCP free-tier with GitHub Actions failover
Agentic Blog Pipeline - 9-Stage Gemini Workflow architecture diagram
Click to enlarge
PythonGoogle GeminiSanity CMSGCP Cloud SchedulerGitHub ActionsSERP / PAA grounding

AI-Powered Portfolio with Generative UI

Interactive portfolio with Gemini-powered chatbot, multi-layer security (26 penetration vectors blocked), and hybrid context architecture (70% token reduction)
AI-Powered Portfolio with Generative UI architecture diagram
Click to enlarge
Next.js 15React 19TypeScriptGoogle Gemini 2.5 FlashVercel AI SDK 5.xZodTailwind CSSFramer Motion

Lumina - B2B AI Wellness Platform with Computer Vision

17-module eye strain detection platform with MediaPipe, offline-first Electron app, multi-tenant RLS, and Turborepo monorepo architecture
Lumina - B2B AI Wellness Platform with Computer Vision architecture diagram
Click to enlarge
ElectronReactTypeScriptMediaPipeSupabaseNext.jsTurborepoSQLiteCloudflare R2

Technologies

Grab and move around the nodes to explore different technologies I work with.

Education & Activities

Indian Institute of Information Technology, Pune

B.Tech. in Computer Science & Engineering · CGPA 8.14/10

Key Courses: C++, Java, Python, Statistics, DSA, DBMS, OOP, ML, Cloud Computing, Big Data, HPC & Distributed Computing

Research Paper: Deep-learning model with 28 dB average PSNR improvement & 36% OCR improvement

Preview:

Hackathon Achievement

4th place (western region) in "Solving For India" hackathon (2,000+ teams). Built blockchain health-record NFT project; selected by Google to share experience.

Let's Connect

Ready to discuss automation, AI, or potential collaborations? Schedule a meeting or reach out through your preferred channel.

Blogs

My latest articles on software development, AI, and productivity.

Loading...