Shivam Garg
Software Engineer
I'm currently working at Scowtt (Series A) as a Founding Engineer.
On the web
Selected Experience
Founding Engineer
at ScowttAugust 2025 – Present
- •Architected and productionized config-driven general Vertex AI pipelines, replacing FDE-style per-client custom code with standardized deployments for hourly data ingestion, online inference, and weekly retraining—cutting enterprise customer onboarding time from 4 weeks to 2 and enabling parallel multi-client launches.
- •Built and deployed lead-value prediction models using GA4 and CRM features, streaming real-time scores via Cloud Run into Meta and Google Ads Conversion APIs for automated bid optimization.
- •Worked directly with two enterprise clients to translate ad optimization goals into reusable ML systems, driving +59% ROAS in Search and +80% in PMax and supporting Scowtt’s $60M Series A.
Software Engineer
at AGI LabsJuly 2025 – Present
- •Built an automated payments and invoicing system via the Stripe API, saving 15–20 hrs/week in manual work for our client, MediaTeam, a marketing agency.
- •Developed a scalable ad-tech analytics pipeline pulling spend/ROI data from Google Ads, Meta, and TikTok APIs; providing MediaTeam with dashboards with <1% sync failure and 5-min data freshness.
Software Engineer and Co-founder
at SpindlNovember 2021 – May 2022
- •Launched multi-chain crypto checkout platform with NFT showcase features, enabling simple link-based transfers and embeddable merchant widgets. (React, Node.js, MongoDB, AWS)
- •Deployed with EC2 Auto Scaling + Docker, CI/CD pipelines, and monitoring via Prometheus/Grafana.
- •Drove adoption to 10k+ users by pairing technical execution with early viral growth strategies.
Selected Writing
Selected Projects
LLM-Powered Podcast Clipping
⭐Clippy transforms long podcasts into 10–15 punchy 30–60s clips by combining WhisperX transcription with an LLM that selects the most engaging Q&A and story moments. Active-speaker detection keeps faces perfectly framed while captions are auto-generated and burned in for vertical, social-ready output. A cloud pipeline—S3 storage, Inngest queues, and Modal GPU workers—handles the heavy lifting asynchronously, and a Next.js web app lets users upload, track progress, preview, and download their clips..
Distributed Lock Service
⭐Fault-tolerant distributed lock manager in Python on top of etcd cluster with reentrant locks.
Recode
⭐An AI DevOps Agent that eliminates downtime by giving your application the ability to fix itself (it does this by automatically detecting errors in production, reading logs, coming up with a fix and deploying it). .
PatchPilot
⭐An autonomous AI engineer/agent that automatically fixes bugs and adds features whenever a GitHub issue is raised, powered by Daytona sandboxes..
ImmunAI
⭐An adversarial security platform powered by an autonomous attack agent that red-teams production AI agents. It autonomously discovers vulnerabilities, builds exploits, and generates novel attack vectors for agents in real time..
Streamling
⭐Decentralized video streaming platform with creator tokenization using DeSo blockchain.
Healera: IoT Faucet Extension
⭐IoT faucet extension that modernizes outdated taps with smart water management system.
TestPrep International
⭐Expanded instructor team to 50+ and student body to 200+ from 10+ countries.