RAGDAIRY — Vocalkisan Dairy AI Document Assistant

A toy RAG (Retrieval-Augmented Generation) demo built to show the dairy sector (NDDB, private dairies, etc.) how their existing website — viksitdairy2047.in — can offer an AI assistant grounded in their own project documents (PDFs, Excel models), via either:

• Option A — Open-source / self-hosted: Ollama running llama3.2:8b locally or on your own server. Full data control, no per-query cost, but needs your own compute and cannot run on free public cloud hosts.
• Option B — Claude (hosted): Anthropic's claude-haiku-4-5 via API. Deployable for free on Streamlit Community Cloud, better answer quality, small per-query cost.

Both options share the same retrieval pipeline (chunking → embeddings → ChromaDB vector search) — only the final "LLM answer generation" step differs. Switch between them from the sidebar radio button in the app.

---

1. Local setup

git clone https://github.com/FGCONSULT/RAGDAIRY.git
cd RAGDAIRY

# reuse or create a conda/venv environment
conda create -n ragdairy python=3.11 -y
conda activate ragdairy

pip install -r requirements.txt

Add a couple of sample documents to data/ (see data/README.md).

To test with Claude (Option B) locally

cp .streamlit/secrets.toml.example .streamlit/secrets.toml
# edit .streamlit/secrets.toml and paste your real ANTHROPIC_API_KEY
streamlit run app.py

To test with Ollama (Option A) locally

ollama pull llama3.2:8b
ollama pull nomic-embed-text   # not required by this app (we use a local
                                # HuggingFace embedding model instead), but
                                # useful if you extend the app later
streamlit run app.py
# In the sidebar, choose "Open-source (Ollama, self-hosted only)"

Open the local URL Streamlit prints (usually http://localhost:8501), upload a PDF/Excel file (or tick "use preloaded sample documents"), click Build / Rebuild Index, then ask a question in the chat box.

---

2. Deploy the free, publicly-hosted version (Option B — Claude)

Streamlit Community Cloud cannot run Ollama, so the public demo link should use the Claude backend.

1. Push this repo to GitHub (already at FGCONSULT/RAGDAIRY) — see §4 below if you're pushing local changes.
2. Go to share.streamlit.io → sign in with GitHub → New app.
3. Repository: FGCONSULT/RAGDAIRY, branch: main, main file: app.py.
4. Click Advanced settings → Secrets and paste:

   ANTHROPIC_API_KEY = "sk-ant-xxxxxxxxxxxxxxxxxxxxxxxx"

5. Deploy. You'll get a public URL such as: https://ragdairy.streamlit.app

Deploying Option A (Ollama) for a persistent demo

Streamlit Cloud won't work for this — instead run it on an in-house server or VPS that has Ollama installed, then either:

• expose it via a subdomain (e.g. rag.vocalkisan.in) behind Nginx, or
• for a one-off live demo from your laptop, run ngrok http 8501 to get a temporary public URL.

---

3. Add it to viksitdairy2047.in (Wix)

Method 1 — Menu link (same pattern as your existing "Dairy Forecast App" link): Wix Editor → Menu → Add Item → paste your Streamlit/subdomain URL → save. Zero coding, matches what's already live on the site.

Method 2 — Inline embed (stays on your domain, feels native):

1. Wix Editor → Add Elements → Embed → Embed a Widget (may show as "Custom Element" / "HTML iframe" depending on your Wix plan).
2. Paste:

   <iframe src="https://ragdairy.streamlit.app/?embed=true"
           style="width:100%; height:800px; border:none;">
   </iframe>

3. Place it on a new page, e.g. "Dairy AI Assistant," alongside your existing Portfolio/Blog pages.

---

4. Pushing local changes to FGCONSULT/RAGDAIRY

git init                      # only if this folder isn't already a git repo
git remote add origin https://github.com/FGCONSULT/RAGDAIRY.git   # if not already set
git add .
git commit -m "Add toy RAG demo app (Claude + Ollama backends)"
git branch -M main
git push -u origin main

If the repo already has content (e.g. an initial README from GitHub), pull first to avoid conflicts:

git pull origin main --allow-unrelated-histories
# resolve any conflicts, then
git push origin main

---

5. What this demo is meant to show

	Option A (Open-source)	Option B (Claude)
Where the model runs	Your own server/laptop	Anthropic's cloud (API)
Data leaves your infrastructure?	No	Only the final prompt + retrieved chunks, per query
Cost	Free (your compute)	Small per-query cost (Haiku-class pricing)
Setup effort	Higher (GPU/server, Ollama)	Lower (just an API key)
Best pitch to	Orgs wanting full data sovereignty (e.g. NDDB internal)	Orgs wanting a fast, low-maintenance pilot

Both point to the same conclusion for the audience: the dairy sector can stand up its own in-house RAG assistant — either fully self-hosted, or as a lightweight hosted pilot — using only their own project documents.

---

Notes / caveats for this toy demo

• Documents uploaded via the app are processed in-memory for that browser session only — nothing is permanently stored, and the vector index is not shared between visitors.
• The preloaded data/ folder should contain only non-sensitive sample files if this repo stays public.
• The guardrail system prompt instructs the model to say "I don't have this information in the uploaded documents" rather than guess — worth demonstrating live by asking an out-of-scope question.