Built at TechX

Real projects, built by TechX teams.

12 projects from our mentor-led teams: AI agents, LLM apps, data science and health AI. Each one was planned, built and presented by students and early-career engineers working with engineers from Google, Microsoft and Meta.

  • AgentStore catalog of AI agents with ratings
    AI Agents

    AgentStore

    An app store for AI agents, where agents can be discovered, installed, rated and remixed.

    What the team built

    • Agent and tool registries with a catalog people can browse
    • An agent runner that records each tool call
    • Ratings, reviews and ranking for trending agents
    • React
    • TypeScript
    • FastAPI
    • MCP-style tools
    • Recommendation models
  • Startup idea validation report with scores for demand, competition and market size
    AI Agents

    Startup Idea Validator

    An AI agent that checks a startup idea and returns a structured validation report.

    What the team built

    • An agent loop that reasons, picks a tool and repeats until done
    • Tools for web search, competitor finding and market sizing
    • Memory of past validations using embeddings
    • Python
    • FastAPI
    • LLM APIs
    • Vector database
  • Inbox assistant with labeled emails and a chat answering calendar questions
    AI Agents

    Inbox & Calendar Assistant

    A chat assistant that searches email by meaning, labels what needs a reply, and manages the calendar.

    What the team built

    • Three cooperating agents: chat, email and calendar
    • Semantic search over thousands of emails
    • Automatic labels such as "needs reply" and "action required"
    • Python
    • LangGraph
    • Embeddings
    • Gmail and Calendar APIs
  • Weekly medical research digest ranked by interest score
    Health AI

    MedDigest

    A research digest that collects new medical papers, summarizes them with AI and emails a newsletter.

    What the team built

    • Automatic paper collection and classification by specialty
    • AI-generated interest scores and summaries
    • Web app with sign-up and an email newsletter
    • Next.js
    • FastAPI
    • LLM APIs
    • Firestore
    • Gmail API
  • Retinal image with a heatmap and predicted condition probabilities
    Health AI

    Retinoscope

    An explainable deep learning model that screens retinal images for seven conditions.

    What the team built

    • Transfer learning on ResNet-50 with texture features
    • Heatmaps that show which region drove each prediction
    • Plain-English explanations of the result
    • Python
    • PyTorch
    • ResNet-50
    • Grad-CAM
    Watch the demo of Retinoscope (opens YouTube)
  • An AI conversation with a panel to save and share it by link
    LLM Apps

    AI Archives

    A browser extension and web app to save and share ChatGPT, Claude, Gemini and Meta AI conversations by link.

    What the team built

    • Extractors for several AI chat sites
    • A shared format so every conversation looks the same
    • Shareable conversation pages
    • Next.js
    • TypeScript
    • Chrome extension
    • Database
    Watch the demo of AI Archives (opens YouTube)
  • Chat request for a Kyoto trip and a three-day itinerary with weather
    LLM Apps

    Tour Planner

    A travel planner that turns a plain-English request into a day-by-day itinerary.

    What the team built

    • Live data for attractions, hotels, restaurants and weather
    • Itineraries that change through conversation
    • Ranking by budget, pace and interests
    • Python
    • Streamlit
    • LLM APIs
    • Travel and weather APIs
    Watch the demo of Tour Planner (opens YouTube)
  • Chat answering an HR policy question with cited sources
    LLM Apps

    Team Docs Q&A

    A chatbot that answers questions from a team’s Notion and Confluence pages, with sources.

    What the team built

    • Document ingestion and chunking pipeline
    • Retrieval-augmented answers with citations
    • Designed for use inside Slack
    • Python
    • AWS Bedrock
    • Milvus
    • FastAPI
  • A claim marked likely false with a short explanation
    LLM Apps

    AI for Disinformation

    A tool that checks claims against trusted sources and flags AI-generated images.

    What the team built

    • Claim verification with web search
    • Deepfake image detection model
    • Audio and image text extraction so any format can be checked
    • Python
    • Gradio
    • LLM APIs
    • XceptionNet
    • OCR
  • Inbox emails ranked by urgency
    LLM Apps

    ToneRank

    A daily report that ranks the last 24 hours of email by urgency, people first.

    What the team built

    • Separates email from people and from organizations
    • Ranks each group by urgency with an open-source LLM
    • Sends a short priority report
    • Python
    • Gmail API
    • Llama 3
  • Recommended products and a chart of ad uplift by customer segment
    Data Science

    Recommendations & Ad Uplift

    Product recommendations plus ad targeting that spends only where an ad changes behavior.

    What the team built

    • Uplift models that find customers an ad will persuade
    • Calibrated predictions for ad decisions
    • Design for similar-item, co-purchase and banner recommendations
    • Python
    • scikit-learn
    • Uplift modeling
    • Jupyter
  • Chat about monthly spending next to a budget breakdown
    Data Science

    Personal Finance Assistant

    A finance chatbot that combines classic machine learning with a conversational interface.

    What the team built

    • Data preparation for spending categories
    • Planned models for unusual spending and savings forecasts
    • Chat interface on top of the model outputs
    • Python
    • Pandas
    • scikit-learn
    • XGBoost
    • LLM APIs
Icon
TechX