Debug ⚡
FPS 0
Nodes 0
Render 0ms

SrishtiAgarwal

Founding / Software Engineer II

I build backend systems for AI/LLM infrastructure, with a focus on reliability and safety guardrails for production LLM applications. Core contributor to GrowthZ.ai, where I unify ad creative generation, campaign automation, and performance analytics into a single platform for B2B marketing teams. Currently shipping REST APIs, RAG pipelines, and real-time bid optimization.

summary

Founding Engineer building backend systems for AI/LLM infrastructure and production deployment, with 2 years of experience shipping at a fast-moving AI startup.

Jul 2024 - Oct 2024GeniusAI ad copies generator
Nov 2024 - Jan 2025RainbowAI image creatives generator
Feb 2025 - Apr 2025OpusAI video creatives generator
May 2025 - Jul 2025WildcardCampaign management platform
Jul 2025 - Aug 2025Google Ads AuditAuditing of Google Ads campaigns
Sep 2025 - Dec 2025ThriveCampaign optimization platform
Jan 2026 - Feb 2026HawkeyeUnified reporting across channels
Mar 2026 - Jun 2026Multi-Channel IntegrationMulti-channel data aggregation across GrowthZ.ai
Jul 2026 - Aug 2026Video-to-Meta-Ad PipelineVideo-to-ad pipeline for Meta's top spenders

experience

Founding Engineer · Full Time

@ GrowthZ AI · July 2024 - Present
  • Core contributor to GrowthZ.ai: an AI platform that unifies ad creative generation, campaign automation, and performance analytics for B2B clients, from system architecture to production deployment.
  • Designed scalable REST APIs across 6 microservices powering a multi-tenant campaign automation engine that auto-generates keywords, ad groups, creatives, and extensions; trained and tuned the LightGBM model driving real-time bid optimization.
  • Built an LLM-powered ad copy generation pipeline that scrapes Play Store/App Store/website data, extracts sentiment and themes, and generates structured, keyword-researched ad copies with a RAG workflow, embedding brand content with Gemini embeddings and Google Search grounding for context-aware ad group and phrase generation, producing static, animated, and AI-generated video creatives at sub-60s latency.
  • Architected an automated video-to-Meta-ad pipeline for one of India's largest Meta advertisers, processing ~1,000 video uploads/day across 60+ ad accounts. Moved uploads to direct-to-object-storage streaming, cutting cloud egress costs ~60%.
  • Maintained AWS infrastructure with Docker and Pulumi (IaC), cutting costs 30% via NAT optimization, resource rightsizing, and scheduled shutdowns, while sustaining 99% uptime.
  • Built a performance analytics layer aggregating multi-platform campaign data into a unified marketing dashboard, surfacing real-time performance signals to the marketing team.
  • Designed and built Creative Suite, an AI ad-copy-to-creative generator, as sole engineer end-to-end. Adopted by 9 clients within 3 months, opening a new monetization channel. Recognized as Top Performer for 3 consecutive months (140% of quarterly variable).
  • Independently built a Google Ads audit service running performance & hygiene checks across account categories, cutting manual audit time from 3-4 hours to under 3 minutes per account, accelerating client onboarding.

projects

LLM SentinelLLM Observability & Governance Platform
Live
  • Built a drop-in reverse proxy between any app and an LLM provider, running synchronous pre-call guardrails without adding latency to the client's request path.
  • Designed async LLM-as-judge quality scoring (hallucination, toxicity, faithfulness) off the critical path, feeding a tamper-evident, hash-chained audit log with EU AI Act Article 12-compliant reporting.
  • Implemented provider fallback with per-provider circuit breakers, measured at p50 ~29ms / p95 ~46ms gateway overhead under load.
NestJSTypeScriptGeminiRedisPostgreSQLSQS
CallSense AIAI Call-Intelligence Platform
Live
  • Built a transcription pipeline with a 3-provider chain (Whisper, Smallest Pulse, Sarvam) and real speaker diarization using VAD + voice embeddings + clustering, not heuristic speaker alternation.
  • Architected LLM rubric-based QA scoring using schema-validated structured outputs, plus sentiment analysis, keyword topic tracking, and semantic search via Gemini embeddings + cosine similarity.
  • Streamed live transcription job status via Server-Sent Events, with each pipeline stage degrading gracefully when its provider dependency is unavailable.
FastAPI (Python)Next.jsGeminiWebRTC VAD

skills

Languages

TypeScriptJavaScriptPythonSQLC++

AI / ML

LangChainLangGraphGemini Vision AIRAGLLM-as-Judge EvaluationGemini 2.5 FlashEmbeddingsImagen 3Veo 3

Frameworks & Libraries

Node.jsNestJSExpress.jsFastAPI

Databases & Systems

PostgreSQLMongoDBMySQLDuckDBSystem Design (HLD/LLD)DSAMicroservicesGit

Infra & Cloud

AWS (S3, SQS, Lambda, EC2, API Gateway)GCP (Cloud Run, Cloud SQL)Cloudflare R2DockerPulumi (IaC)

leetcode

Loading...

education

B.Tech in Computer Science and Engineering

@ SRM Institute of Science and Technology · Sept 2020 - Jun 2024

Graduated with 9.25 CGPA.

contact

Open to interesting problems, collaborations, and conversations about backend systems and AI or just tell me your bid optimizer is on fire, I'll relate either way.