Harsh Gidwani
AI Engineer building production-grade multi-agent systems, LLM applications, and durable execution pipelines
about
Harsh is an AI Engineer who designs and ships production LLM systems — multi-agent pipelines,
retrieval-grounded generation, and durable long-running workflows. His flagship product, Lemma, runs a
five-agent analysis pipeline in production: adversarial critique agents audit every output,
market figures must cite live-retrieved sources, and the whole workflow survives failures through
step-level retries on serverless infrastructure.
He works across the full stack — Next.js, TypeScript, and PostgreSQL on the product side;
Gemini, LangChain, and schema-constrained generation on the AI side — and cares most about the hard
middle: making LLM systems reliable, grounded, and observable enough to trust in production. He is
completing an Integrated B.Tech in AI & Data Science at MIT-WPU alongside professional AI
engineering work.
skills
AI Engineering
- Multi-Agent Orchestration
- Durable Execution (Inngest)
- Retrieval Pipelines / RAG
- LLM Evaluation & Grounding
- Prompt Engineering
- Fine-Tuning
Languages & Frameworks
- Python
- TypeScript
- Next.js (App Router)
- LangChain
- TensorFlow
- Scikit-Learn
Infrastructure & Tools
- PostgreSQL + Prisma
- Gemini API
- Serverless (Vercel)
- Redis / Rate Limiting
- Git & CI
- Zod / Schema Validation
Hiring for an AI engineering role?
See the systems I've shipped — architecture decisions, trade-offs, and results — or grab the resume for the short version.
projects
Lemma
Production multi-agent platform — five grounded AI agents evaluate a research paper's commercial potential and generate a fully-cited investor deck, on durable serverless workflows.
GrowAI
End-to-end AI marketing automation platform — fine-tuned LLM pipeline that converts raw product data into multi-channel ad scripts, emails & social copy.
Finura
Smart financial planning & budget intelligence system — LSTM forecasting, AES-256 encrypted storage, and real-time spending anomaly detection.
CognisCRM
AI-driven customer lifecycle & business intelligence suite — Llama 3.1 multimodal lead intelligence, predictive scoring, and enterprise-scale microservices.
LCEL Chatbot
Advanced AI chatbot using LangChain Expression Language with context-aware responses, memory retention, and real-time streaming.
BERT Visualization
Interactive tool visualizing BERT's attention patterns, token embeddings, and layer representations for NLP explainability.
Customer Churn Analysis
ML solution predicting churn probability — XGBoost, feature importance, and 89% accuracy with 0.92 AUC on validation data.
YouTube Summarizer
GenAI tool extracting and summarizing YouTube video transcripts — supports multiple languages and customizable summary lengths.
NLP Resume Parser
Automated resume data extraction using spaCy — skills, experience, education across PDF, DOCX for efficient recruitment workflows.
OpenAI Chatbot
Multi-turn conversational chatbot powered by OpenAI APIs with contextual awareness and natural language generation via Flask.
Ollama LLM Chatbot
Fully private local LLM chatbot using Ollama — zero external API calls, runs entirely on-device for maximum data privacy.
IMDB Sentiment Analysis
Sentiment analysis pipeline comparing SVM, Naive Bayes, and LSTM models on IMDB reviews with rich visualization output.
