Bengaluru, IN · 12.97°N 77.59°E2026 · Issue 01

    § Founding AI / ML Engineer  ·  Video Generation  ·  Multi-Agent Production Systems

    The summary

    Founding AI/ML engineer at JuiceLabs. I built our video generation pipeline. Idea goes in, finished cinematic video comes out. Length-agnostic. Currently shipping ads at millions of impressions a month for enterprise clients. Earlier I was the founding engineer at Sytolab, where I built the human-machine interface for a biomedical 3D printer used by researchers for organ regeneration studies. Based in Bengaluru, India.

    dhruvpuri.cloud

    §01 · Who

    About

    I'm a founding AI/ML engineer at JuiceLabs. Before that I was the founding software engineer at a biotech startup building 3D bioprinters. I've done 0-to-1 twice now, both times without a playbook.

    At JuiceLabs I built the AI pipeline from scratch. Prompt strategies across Sonnet, Gemini, and VEO. Agent workflows. Automated evaluation loops. The inference infrastructure underneath. The pipelines generate ad creatives for enterprise brands, and those ads get seen by millions of viewers. I also run weekly Claude Code sessions for our 20-person team (four months in) and consult directly with clients on AI adoption.

    At Sytolab I built Syto Studio, the HMI researchers used to 3D-print tissue constructs for organ regeneration studies. React and Flask on the surface, Klipper and Moonraker for printer control, OpenCV for real-time print quality. The work sounds dry written down. It wasn't. The output went into living animals.

    Earlier I was an SDET at Paytm, automating device testing for systems that processed 10M+ transactions. Before that, product analyst at Terribly Tiny Tales. I cold-DMed TTT as a first-year ECE student, offered to build their Discord for free, and turned it into a two-year role. My degree is in Electronics & Communications. Everything CS is self-taught.

    §02 · Work

    Experience

    OCT 2025 - Present

    Bengaluru, India

    Founding AI/ML Engineer

    Juicelabs.ai

    • Built JuiceLabs' video generation pipeline from idea to final video. Currently shipping cinematic ads; pipeline is length-agnostic. Multi-phase agentic architecture with discrete user-approval gates at every expensive step. Multi-model orchestration across Opus 4.6, gpt-image-2, Seedream 4.5, and Seedance 2.0. Cross-family LLM-as-judge evaluation so the generator and grader don't collude. Output ships to enterprise clients at millions of impressions a month.
    • Founding engineer (0→1). Built JuiceLabs' AI pipeline from scratch on FastAPI + Firestore + GCP, with LangFuse for observability, Qdrant for retrieval, and direct SDK integrations with OpenAI, Anthropic, and Google. Parallelism via asyncio.gather across the agent graph.
    • Run weekly Claude Code sessions for our 20-person team. Four months and counting. Cover prompt strategies, agentic workflow design, and how to integrate Claude into actual day-to-day engineering work.
    • Advise enterprise clients directly on AI adoption and Claude integration as part of production deliverables. Forward-deployed AI work in everything but title.

    JAN 2024 - OCT 2025

    Bengaluru, India

    Founding Full Stack AI Engineer

    SYTOLAB/KREATOR3D

    • Built Syto Studio, the human-machine interface for biomedical 3D printers. React frontend, Flask backend. Integrated Klipper for STL uploads, slicing, and temperature controls; Moonraker for real-time printer monitoring.
    • Researchers used the system to print tissue constructs placed into living animals for organ regeneration studies.
    • Wrote the OpenCV-based computer vision layer for real-time print quality control. Reduced user setup time by 30% and improved print accuracy by 50%.

    JAN 2023 - JAN 2024

    Delhi, India

    Software Development Engineer in Test

    PAYTM

    • Automated device testing using Python, reducing testing cycles by 50% and eliminating 25 hours/week of manual effort
    • Authored 100+ manual test cases for Soundboxes, covering functional, regression, and integration scenarios
    • Created comprehensive test plans and maintained documentation for end-to-end testing cycles
    • Integrated AWS services (S3, DynamoDB) with Boto3 SDK to streamline test data management
    • Generated Matplotlib-powered test reports, identifying 50+ critical bugs pre-release and cutting post-release issues by 20%
    • Tested REST APIs for high-volume transactional systems, ensuring data integrity across 10M+ network requests

    OCT 2021 - JAN 2023

    Mumbai, India

    Product Analyst

    TERRIBLY TINY TALES(TTT)

    • Revamped frontend UI/UX using React and Cypress, boosting website traffic by 20% and reducing production bugs by 40%
    • Led cross-functional collaboration with UX designers and product managers to align and prioritize features
    • Collaborated with DevOps to streamline microservices testing in Docker containers, improving deployment reliability
    • Performed cross-browser testing (Chrome, Firefox, Safari) for React-based web applications, resolving 30+ UI compatibility issues
    • Automated regression testing with Cypress, increasing test coverage by 35%

    §03 · Work shipped

    Selected projects

    Project 00

    Video Generation Pipeline

    JuiceLabs' flagship. Idea goes in, finished cinematic video comes out. Multi-phase agentic architecture with multi-model orchestration and cross-family LLM-as-judge evaluation. Shipping ads at millions of impressions a month.

    FastAPIGCPFirestoreOpus 4.6gpt-image-2Seedream 4.5Seedance 2.0LangFuse
    Read case study
    Video Generation Pipeline

    Project 00

    Parameter Golf Autoresearch

    OpenAI Model Craft Challenge: built an autonomous research agent that runs hyperparameter sweeps, scores candidates against the frontier, and decides what to test next. RTX 4060 for iteration, 8xH100 via RunPod for finals.

    PythonPyTorchCUDARunPodH100RTX 4060Hyperparameter OptimizationSmall LM Training
    Read case study
    Parameter Golf Autoresearch

    Project 00

    Yelp AI Review Classifier

    Fine-tuned DistilBERT plus local Qwen3-8B for zero-cost sentiment classification. 81% accuracy with a domain-shift analysis that mattered more than the headline number.

    PythonDistilBERTQwen3-8BTransformersPyTorchHugging FaceNLPMachine Learning
    Read case studySource
    Yelp AI Review Classifier

    Project 00

    AI Video Ad Concept Analyzer

    GPT-5-powered evaluator that scores ad concepts across six dimensions. Built to test whether LLM feedback on creative work can be useful or just drifts with temperature.

    PythonOpenAI GPT-5FastAPIJSONCLIREST APIPrompt EngineeringPerformance Monitoring
    Read case studySource
    AI Video Ad Concept Analyzer

    Project 00

    Policy Compare

    AI-powered policy comparison that helps regular people understand 50-page loan documents. Built after watching my parents lose hours to bank PDFs.

    ReactTypeScriptStreamlitRenderWeb Technologies
    Read case study
    Policy Compare

    Project 00

    AgentVC — AI Investor Interviewer

    AI-powered pitch practice platform where founders rehearse against a simulated investor. Live demo shipped, insights from 50+ real investor conversations baked into the feedback logic.

    Next.jsTypeScriptOpenAI APIVercelMachine LearningNLPSentiment Analysis
    Read case study
    AgentVC — AI Investor Interviewer

    Also on the shelf

    §04 · References

    In their words

    Working with Dhruv was a game-changer for our project. Their expertise and dedication were invaluable.
    Amar BharadwajBanking Professional
    A true professional who delivers high-quality work. It was a pleasure collaborating with Dhruv.
    Nandini BishtCreative Director, Pixel Panga Studios
    Exceptional problem-solving skills and innovative solutions. Dhruv exceeded our expectations.
    Ojash YadavFounder, Notebook Studios

    §05 · Stack

    Toolbox

    Programming Languages

    Python · TypeScript · JavaScript · Go · Java · SQL · C/C++ · HTML/CSS

    Frameworks & Libraries

    React · Next.js · Node.js · Express.js · Flask · FastAPI · Tailwind CSS

    LLM & AI Models

    Claude (Opus + Sonnet) · GPT (incl. gpt-image-2) · Gemini · Seedream · Seedance · LangGraph · LangChain · Qdrant · LangFuse · vLLM · Ollama · Prompt Engineering · LLM-as-Judge Evals

    ML & Computer Vision

    PyTorch · Transformers · Hugging Face · OpenCV · YOLO · LLM Fine-tuning (LoRA/QLoRA) · Scikit-learn

    Prompt Engineering

    JSON Prompting · XML Prompting · Structured Prompts · Multi-modal Steering · Chain-of-Thought

    Generative AI

    Image Generation · Video Generation (Seedance 2.0) · Seedream 4.5 · Multimodal Pipelines

    Data & Analytics

    Pandas · NumPy · Matplotlib · Seaborn · Kafka · WebSocket

    Workflow Automation

    N8N · Make.com · Agent Orchestration · Automation Pipelines

    Testing

    Pytest · Cypress · Vitest · Selenium · Postman · End-to-End · Regression · Cross-browser

    Cloud & DevOps

    AWS (S3, DynamoDB) · Docker · Kubernetes · Terraform · Jenkins · GitHub Actions · Vercel · Netlify · Supabase · Git

    Databases

    PostgreSQL · MongoDB · DynamoDB · Redis · SQLite · ChromaDB · Qdrant

    Distributed Systems

    Fault-tolerant Architecture · Low-latency Systems · Microservices · Query Optimization

    §06 · Contact

    Say hello

    Interested in new opportunities, short-scope advisory, or a genuine conversation about AI engineering. Send a short note with what you’re working on, your timeline, and what decision you’re trying to make. If it’s a fit I reply within a few business days.