AI ENGINEER · LLMs · AI AGENTS · RAG · PRODUCTION ML
Manan Maroo
I build LLM-powered products, AI agent systems, and production ML pipelines. Founder of Agentshive — an open registry for AI agents. 3+ years turning ambiguous problems into shipped products: 40% faster research turnaround, 20% campaign efficiency gains, live systems used by real clients.
Featured · Founder Project
Agentshive LIVE
An open registry for AI agents supporting Claude Code, Codex, n8n, and LangChain.
Ships a CORS-enabled REST API for one-command agent installs and an MCP server
(@agentshive/mcp) exposing search_agents, get_agent, and
install_agent tools directly inside Claude Code, Cursor, and other MCP clients.
Deep dives
Case Studies
Agentshive — AI Agent Registry
Problem Installing AI agents meant manual copy-paste of prompt files across Claude Code, Codex, n8n, and LangChain — no shared source of truth.
Architecture Next.js 16 + Supabase (Postgres + Auth) on Vercel with edge-cached APIs; a CORS-enabled REST endpoint serving prompts as markdown, plus an @agentshive/mcp server exposing search/get/install tools over stdio.
Impact One-command agent installs via curl, WebFetch, or MCP — eliminating manual setup — in a live open-source registry used across four agent runtimes.
Document Intelligence API
Problem Analysts needed to query large structured datasets (Excel, CSV) in natural language instead of writing bespoke scripts per request.
Architecture A production API that grounds an LLM over structured data with retrieval and output-evaluation frameworks, with systematic prompt engineering to cut error rates.
Impact Natural-language querying adopted across 3+ enterprise client workflows, removing hand-written analysis for routine questions.
LLM Synthetic Personas
Problem Market research depended on slow, expensive live respondent panels to answer client questions.
Architecture An LLM-powered synthetic persona system generating panel-like responses, with evaluation frameworks to keep outputs reliable across accounts.
Impact Cut client research turnaround by 40% across 10+ enterprise accounts, removing dependency on live panels.
Spotify AI DJ
Problem Streaming playlists lack DJ-style transitions and fine control over how tracks are mixed and sequenced.
Architecture Full-stack mix engine (FastAPI + React) over the Spotify and YouTube APIs with configurable mix points, popularity filtering, artist blacklisting, and dynamic queue reordering.
Impact A live, demo-friendly product that turns any library into a continuously mixed set.
Career
Experience
Founder & AI Engineer
Dec 2025 – PresentAgentshive · Remote
- Founded and shipped agentshive.net, an open registry for AI agents across Claude Code, Codex, n8n, and LangChain — Next.js 16, Supabase (Postgres + Auth), Vercel with edge-cached APIs.
- Designed a CORS-enabled REST endpoint serving agent prompt files as markdown, enabling single-command installs via curl, WebFetch, or MCP.
- Built an MCP server (
@agentshive/mcp) in TypeScript exposing search, fetch, and install tools over stdio for MCP-compatible clients. - Curated production-ready agents across code review, security review, test generation, and documentation, with standardised metadata and prompts across runtimes.
AI Engineer
Jul 2025 – Jul 2026Strategic Research Insights · New Jersey, USA
- Engineered an LLM-powered synthetic persona system that cut client research turnaround by 40% across 10+ enterprise accounts.
- Launched a production document intelligence API for natural-language querying of Excel/CSV datasets, adopted across 3+ client workflows.
- Built ETL pipelines (Python, SQL, R, SAS) processing 1M+ rows per run; reduced LLM error rates via systematic prompt engineering and evaluation frameworks.
- Converted ambiguous business questions into statistical models that shaped go-to-market strategy for 3 enterprise clients — findings adopted without revision.
Lead Data Analyst
Jan 2024 – Jun 2025Jain Management & Minds Beyond Measure · New York, USA
- Led analytics across 5 concurrent performance marketing campaigns; audience segmentation framework lifted campaign efficiency by 20%.
- Built real-time Tableau and Google Sheets dashboards, compressing stakeholder reporting from weekly cycles to same-day visibility.
- Delivered prescriptive analytics to executives; recommendations adopted without revision in 4 of 5 quarterly leadership reviews.
Data Science Co-op
Sep 2023 – Dec 2023Bayer · New Jersey, USA
- Processed and validated 3M+ records through Python ETL pipelines for persona-based customer segmentation and behavioral analytics.
- Applied clustering and predictive modeling to surface behavioral patterns used in downstream product targeting.
AI Engineer Intern
Jun 2023 – Aug 2023Boost Money · Fintech
- Built a BERT-based NER model reaching 92% accuracy on financial documents — contracts, statements, regulatory filings.
- Integrated Hugging Face transformers via REST API, cutting manual review time 30% across 50K+ financial records.
Portfolio
Projects
Live Products
Agentshive — AI Agent Registry
Open registry for AI agents with one-command installs and a TypeScript MCP server integrating directly into Claude Code and Cursor.
Supply Chain Market Intelligence
Search any company and map its suppliers, customers, and live stock prices across 10+ global exchanges refreshing every 15 seconds, with AI-generated supply-chain risk analysis.
Spotify AI DJ
Full-stack AI music mix engine for Spotify and YouTube: configurable mix points, popularity filtering, artist blacklisting, and dynamic queue reordering.
AI Tools & NLP
Resume Forge
AI resume generator producing ATS-optimized outputs across 8 country-specific templates, tailoring tone, keyword density, and formatting per region.
Financial NER Model
BERT-based Named Entity Recognition at 92% accuracy on financial documents — automated entity extraction from contracts, statements, and filings.
Custom NLP Entity Recognizer
SpaCy pipeline extracting and visualizing entities from PDFs — people, orgs, locations, dates, and financial figures with inline annotated spans.
AI Content Analyzer & SEO Generator
Web text analyzer for SEO evaluation with RLHF-based feedback loops that refine recommendations through user ratings — 30% quality lift over baseline.
Visualforall — Accessibility Tool
Helps visually impaired users interpret online content through image captioning and OCR, designed with a strong assistive-technology UX focus.
Font Detection App
Detects fonts in images and highlights where font changes occur — useful for design QA, brand consistency, and accessibility testing.
Data & Analytics
Financial Market Prediction Engine
Market forecasting at 85% prediction accuracy using regression and statistical modeling, plus an R-based portfolio allocation optimizer.
Next-Gen Job Recommender
Scalable ML job recommendation engine improving match rates by 40%, built on FastAPI + MongoDB for production-scale throughput.
NBA Performance Intelligence
Interactive dashboards across 20+ NBA seasons surfacing scoring trends, team performance patterns, and player trajectories.
Marketing Analytics Toolkit
Reusable analyst toolkit covering market segmentation, campaign effectiveness scoring, and business review scraping (BeautifulSoup, Scrapingdog API).
Kaggle Competitions
Top-100 finish (rank 64) in Digit Recognition; Predict Student Performance from Game Play using ANN, CNN, CatBoost, and LightGBM ensembles.
Computer Vision & Apps
E Medicare
Medical portal with online prescriptions, report storage, and medication reminders — end-to-end patient convenience platform.
Facial Mood Recognition
Real-time facial recognition detecting 5 distinct moods, built with OpenCV and a Keras expression-recognition model.
Invisibility Cloak
Harry Potter-inspired invisibility effect using real-time image recognition and background substitution with OpenCV2.
Movie Recommendation Engine
Content-based movie recommender using cosine similarity, with facial-recognition integration for mood-aware suggestions.
Publication
Task Scheduling & Load Balancing in Cloud Computing using Firefly Algorithm
Peer-reviewed paper presented at the 3rd International Conference on Computing, Communication and Cyber Security (IC4S 2021).
Toolbox
Skills
GenAI & LLMs
Data Science & ML
Programming
Frameworks
Cloud & Data Engineering
Visualisation & Tools
Background
Education & Certifications
M.S. Data Science
Machine Learning · Statistical Modelling · NLP · Recommender Systems · Experimentation Design · Data Engineering
B.Tech Computer & Communication Engineering
ACM · Robotics Club · Microsoft Student Ambassador
Languages
English (native) · Hindi (native) · Dutch (A1) · German (A1)
- Google Advanced Data Analytics
- Oracle Generative AI Professional
- AWS Cloud Practitioner
- Azure Fundamentals (AZ-900)
- Google Cloud Associate
- SQL for Data Science — UC Davis
- Databases & SQL for Data Science — IBM
- Python for Data Science, AI & Development — IBM
- Predictive Modelling with Azure ML Studio
- Facial Expression Recognition with Keras — Coursera
- Machine Learning Foundations — Univ. of Washington
- How Google Does Machine Learning — Google Cloud
Say hello
Let's build something
Open to AI/ML engineering and data science opportunities worldwide. Happy to talk about LLM products, AI agents, or anything data.