🔗 LinkOps Engine

Automated Startup Ecosystem Matchmaking Dashboard

Built for program administrators at accelerators and national innovation funds.

View Demo

---

About

LinkOps is a human-in-the-loop AI dashboard that automates the process of matching startup pitch decks to the right mentors and ecosystem partners. An admin uploads PDF or image pitch decks, the AI reads and analyses them using Gemini 2.5 Flash, and surfaces a ranked recommendation for the single best mentor and partner match from a curated database — with a 2-sentence justification for each decision.

The admin then approves, rejects, or interrogates those decisions through an Explainable AI chat before any linkage is committed to the ledger. Nothing gets approved without a human saying so.

Built as a demo for Cradle Fund Malaysia, but the architecture is generic enough to drop into any accelerator, corporate VC, or government innovation program.

---

Features

Pitch Deck Processing

• drag-and-drop bulk upload (PDF, PNG, JPG, JPEG)
• up to 5 files per session, 10MB per file
• per-file status badges — Analysed / Pending
• smart submit button — counts only unprocessed files, disables itself when nothing new is queued
• animated progress bar that fills realistically as each deck is analysed (API runs in a background thread)
• granular error handling — partial failures reported individually, success/error toast notifications for every action

AI Matchmaking Engine

• reads the actual pitch deck content (multimodal — not just filename)
• matches against a live CSV database of 30 mentors and 18 partners
• returns structured JSON: startup profile + best mentor match + best partner match, each with a 2-sentence justification
• temperature 0.3 for consistent, low-hallucination output

Human-in-the-Loop Approvals

• each processed startup gets an approval card with its profile, recommended mentor, and recommended partner
• one-click Approve for each linkage — idempotent (can't double-approve)
• dismiss cards once done
• approved linkages flow to the ledger instantly

Explainable AI (XAI) Chat

• per-startup chat modal — ask the AI why it picked a specific mentor or partner
• scoped strictly to matchmaking — refuses off-topic questions with a consistent response
• handles questions about names not in the database (explains they're not in the pool)
• streaming response rendered inside a scrollable container, no layout jumps
• rate limiting: 3s cooldown between messages, 20 message session cap, 500 char input limit
• Malaysian 3R content guardrails at the system instruction level

Ecosystem Pool Browsers

• Browse Mentor Pool / Browse Partner Pool buttons in the sidebar
• opens a full-screen modal with the entire database in a native sortable, filterable table
• scales to any size — no inline sidebar rendering

Approved Linkages Ledger

• custom HTML table — Match Reason column wraps text fully, nothing gets cut off
• status badge, hover highlight, full match justification always visible
• one-click CSV export of the full ledger

Sidebar

• live session metrics (files queued, processed, approved)
• pool browser shortcuts
• CSV export shortcut when linkages exist

---

Tech Stack

Layer	Tool
UI Framework	Streamlit ≥ 1.30
AI Model	Gemini 2.5 Flash via google-genai SDK
Data	Pandas + CSV (mentor/partner/startup databases)
Concurrency	concurrent.futures.ThreadPoolExecutor
Containerisation	Docker
Fonts	Geist (body), Material Symbols Outlined (icons)

---

Prerequisites

• Python 3.11+
• A Google AI Studio API key with Gemini 2.5 Flash access
• pip

---

Installation

git clone https://github.com/YOUR_USERNAME/linkops.git
cd linkops
pip install -r requirements.txt

---

Configuration

Create .streamlit/secrets.toml:

GEMINI_API_KEY = "your_api_key_here"

if deploying via Docker or Cloud Run, set GEMINI_API_KEY as an environment variable instead — the app reads from os.environ first.

---

Running Locally

streamlit run app.py

opens at http://localhost:8501

---

Docker

docker build -t linkops .
docker run -p 8080:8080 -e GEMINI_API_KEY=your_key linkops

---

Project Structure

linkops/
├── app.py                  # ~70 lines — page config, state init, section calls
├── config.py               # constants (MAX_FILES, model name, XAI limits)
├── requirements.txt
├── Dockerfile
├── core/
│   ├── __init__.py
│   └── matchmaker.py       # execute_match_protocol, query_xai, approve_linkage
├── ui/
│   ├── __init__.py
│   ├── styles.py           # inject_styles(), show_toast()
│   ├── dialogs.py          # all @st.dialog functions (pools, preview, XAI chat)
│   ├── sidebar.py          # render_sidebar()
│   ├── file_manager.py     # render_file_manager() + DummyFile
│   ├── approvals.py        # render_approvals()
│   └── ledger.py           # render_ledger()
├── data/
│   ├── mentors.csv         # mentor database (30 mentors)
│   ├── partners.csv        # partner database (18 partners)
│   └── startups.csv        # sample startups for reference
└── .streamlit/
    ├── config.toml         # theme + toolbar config
    └── secrets.toml        # api key (gitignored)

---

Extending the Databases

The mentor and partner pools are flat CSVs — swap them out or append rows, no code changes needed.

mentors.csv columns: Name, Expertise

partners.csv columns: Name, Focus

The AI reads the full database content at inference time, so new entries are picked up immediately on the next session start.

---

Limitations

• session state is in-memory — approved linkages reset on page refresh (no database persistence yet)
• file/size limits are defined in config.py (MAX_FILES, MAX_FILE_SIZE_MB)
• XAI chat history is per-session only

---

License

MIT — do whatever you want with it.

---

Built at a hackathon. Powered by Gemini.