Upload 8 files

README.md

@@ -1,20 +1,28 @@
----
-title: Salesforce Generate Pre Call Summary
-emoji: 🚀
-colorFrom: red
-colorTo: red
-sdk: docker
-app_port: 8501
-tags:
-- streamlit
-pinned: false
-short_description: To generate pre call summary for agent
-license: apache-2.0
----
 
-# Welcome to Streamlit!
+# Pre‑Call Summary Generator (Salesforce → Hugging Face Space)
 
-Edit `/src/streamlit_app.py` to customize this app to your heart's desire. :heart:
+A Streamlit app that:
+1) Accepts Salesforce credentials (instance URL + access token).
+2) Pulls Account, Opportunities, Cases via REST/SOQL.
+3) Calls a summarization **model** (Hugging Face Inference API by default).
+4) Shows a structured **pre‑call summary** and (optionally) writes it back to Salesforce as a Note.
+
+## How to run on Hugging Face Spaces
+
+1. Create a new Space → **SDK: Streamlit** → **Private** (recommended for prototypes).
+2. Upload this repo's files.
+3. In the Space → **Settings → Repository secrets**, add your tokens:
+   - `HF_TOKEN` : (required) A Hugging Face token with Inference API access.
+4. (Optional) For testing locally, you can run `streamlit run app.py`.
+
+## Security Notes
+- Do **not** log or persist Salesforce tokens. This app keeps everything in memory only for the session.
+- For production, prefer Salesforce OAuth Web Server flow instead of pasting tokens.
+- Scope the SOQL fields and follow FLS/sharing best practices in your org.
+
+## Usage
+- Enter `instance_url` (e.g., `https://yourDomain.my.salesforce.com`) and a valid **Access Token**.
+- Provide an `Account Id` (e.g., starts with `001...`).
+- Click **Generate Summary**.
+- Optionally check **"Write to Salesforce as Note"** to save the JSON summary back.
 
-If you have any questions, checkout our [documentation](https://docs.streamlit.io) and [community
-forums](https://discuss.streamlit.io).

app.py

@@ -0,0 +1,71 @@
+
+import streamlit as st
+import json
+from salesforce import get_context, create_note
+from model import PreCallSummaryModel
+from schema import PreCallSummary
+
+st.set_page_config(page_title="Pre‑Call Summary Generator", layout="wide")
+
+st.title("Pre‑Call Summary Generator (Salesforce → HF Space)")
+
+with st.expander("Instructions", expanded=False):
+    st.markdown("""
+    1. Enter your Salesforce **Instance URL** (e.g., `https://xxx.my.salesforce.com`).
+    2. Provide a short‑lived **Access Token** (from a Connected App / OAuth flow).
+    3. Provide an **Account Id** (`001...`).
+    4. Click **Generate Summary**. Optionally save it as a **Note** in Salesforce.
+    **Security:** tokens are used only in memory for this session and are not logged.
+    """)
+
+inst = st.text_input("Salesforce Instance URL", placeholder="https://yourDomain.my.salesforce.com")
+token = st.text_input("Salesforce Access Token", type="password", placeholder="Paste a short‑lived token")
+acct_id = st.text_input("Account Id", placeholder="001xxxxxxxxxxxx")
+
+colA, colB, colC = st.columns([1,1,1])
+with colA:
+    opp_n = st.number_input("Max Opportunities", min_value=0, max_value=20, value=5, step=1)
+with colB:
+    case_n = st.number_input("Max Cases", min_value=0, max_value=20, value=5, step=1)
+with colC:
+    task_n = st.number_input("Max Tasks", min_value=0, max_value=50, value=10, step=1)
+
+push_back = st.checkbox("Write to Salesforce as Note after generation", value=False)
+generate = st.button("Generate Summary", type="primary")
+
+if generate:
+    if not inst or not token or not acct_id:
+        st.error("Please provide instance URL, access token, and account id.")
+        st.stop()
+
+    with st.spinner("Fetching Salesforce context..."):
+        try:
+            context = get_context(inst, token, acct_id, opp_n, case_n, task_n)
+        except Exception as e:
+            st.error(f"Salesforce API error: {e}")
+            st.stop()
+
+    st.subheader("Fetched Context (trimmed)")
+    st.json(context)
+
+    with st.spinner("Calling model..."):
+        model = PreCallSummaryModel()
+        try:
+            summary: PreCallSummary = model.generate(context)
+        except Exception as e:
+            st.error(f"Model error: {e}")
+            st.stop()
+
+    st.subheader("Pre‑Call Summary (JSON)")
+    st.json(json.loads(summary.model_dump_json()))
+
+    if push_back:
+        with st.spinner("Writing Note to Salesforce..."):
+            try:
+                note_id = create_note(inst, token, acct_id, "Pre‑Call Summary", summary.model_dump_json(indent=2))
+                st.success(f"Saved Note Id: {note_id}")
+            except Exception as e:
+                st.error(f"Failed to save Note: {e}")
+
+    st.download_button("Download JSON", data=summary.model_dump_json(indent=2),
+                       file_name=f"precall_summary_{acct_id}.json", mime="application/json")

model.py

@@ -0,0 +1,53 @@
+
+import json, orjson, json5, requests, os
+from typing import Dict, Any
+from schema import ensure_schema, PreCallSummary
+from prompts import build_prompt
+
+class PreCallSummaryModel:
+    def __init__(self, inference_model_url: str | None = None, hf_token: str | None = None):
+        # If inference_model_url is None, we'll default to FLAN-T5-base via Inference API
+        self.inference_model_url = inference_model_url or             "https://api-inference.huggingface.co/models/google/flan-t5-base"
+        self.hf_token = hf_token or os.environ.get("HF_TOKEN")
+
+    def _call_inference_api(self, prompt: str) -> str:
+        headers = {"Authorization": f"Bearer {self.hf_token}"} if self.hf_token else {}
+        payload = {"inputs": prompt, "parameters": {"max_new_tokens": 512, "return_full_text": False}}
+        r = requests.post(self.inference_model_url, headers=headers, json=payload, timeout=120)
+        r.raise_for_status()
+        data = r.json()
+        if isinstance(data, list) and data and "generated_text" in data[0]:
+            return data[0]["generated_text"]
+        if isinstance(data, dict) and "generated_text" in data:
+            return data["generated_text"]
+        # Fallback: some endpoints return plain text
+        return str(data)
+
+    def _parse_json(self, s: str) -> Dict[str, Any]:
+        # Try strict json, then orjson, then json5, then best-effort substring extraction
+        try:
+            return json.loads(s)
+        except Exception:
+            pass
+        try:
+            return orjson.loads(s)
+        except Exception:
+            pass
+        try:
+            return json5.loads(s)
+        except Exception:
+            pass
+        # naive extraction between first { and last }
+        try:
+            start, end = s.find("{"), s.rfind("}")
+            if start != -1 and end != -1 and end > start:
+                return json.loads(s[start:end+1])
+        except Exception:
+            pass
+        return {}
+
+    def generate(self, context: Dict[str, Any]) -> PreCallSummary:
+        prompt = build_prompt(json.dumps(context, ensure_ascii=False, default=str))
+        raw = self._call_inference_api(prompt)
+        parsed = self._parse_json(raw)
+        return ensure_schema(parsed)

prompts.py

@@ -0,0 +1,66 @@
+
+SCHEMA_TEXT = """
+Return JSON with EXACTLY these keys:
+{
+  "client_overview": str,
+  "relationship_timeline": [
+    {"date":"YYYY-MM-DD","event":"...", "source":"Case|Opportunity|Task|Activity|Other"}
+  ],
+  "open_items": [str],
+  "opportunities": [
+    {"name":"...", "stage":"...", "amount": number, "closeDate":"YYYY-MM-DD or null", "nextStep":"..."}
+  ],
+  "risks_flags": [str],
+  "recommendations": [str],
+  "pre_call_questions": [str],
+  "sources": [str]
+}
+"""
+
+FEW_SHOT = [
+    {
+        "context": {
+            "account": {"Id":"001X","Name":"Acme Corp","Industry":"Manufacturing","Type":"Customer-Direct","Rating":"Hot","AnnualRevenue": 45000000},
+            "opportunities": [
+                {"Id":"006A","Name":"Q4 Renewal","StageName":"Negotiation","Amount": 2000000, "CloseDate":"2025-09-30", "NextStep":"Legal review"},
+                {"Id":"006B","Name":"Cross-sell Treasury","StageName":"Qualification","Amount": 500000, "CloseDate":"2025-10-15", "NextStep":"Intro call"}
+            ],
+            "cases": [
+                {"Id":"500C","CaseNumber":"000123","Subject":"Payment delay","Status":"In Progress","Priority":"High","Origin":"Phone","LastModifiedDate":"2025-08-01T12:00:00Z"}
+            ],
+            "activities": [
+                {"Id":"00T1","Subject":"Quarterly check-in","ActivityDate":"2025-07-15","Status":"Completed","Owner":{"Alias":"jdoe"}}
+            ]
+        },
+        "output": {
+          "client_overview": "Acme Corp (Manufacturing, Customer-Direct). Est. revenue $45M. Relationship rated Hot.",
+          "relationship_timeline": [
+            {"date":"2025-08-01","event":"High-priority case: Payment delay is in progress.","source":"Case"},
+            {"date":"2025-07-15","event":"Completed quarterly check-in.","source":"Activity"}
+          ],
+          "open_items": ["Follow up on payment delay case"],
+          "opportunities": [
+            {"name":"Q4 Renewal","stage":"Negotiation","amount":2000000,"closeDate":"2025-09-30","nextStep":"Legal review"},
+            {"name":"Cross-sell Treasury","stage":"Qualification","amount":500000,"closeDate":"2025-10-15","nextStep":"Intro call"}
+          ],
+          "risks_flags": ["Open high-priority service issue may affect renewal"],
+          "recommendations": ["Confirm payment issue resolution","Schedule renewal legal review","Propose treasury intro"],
+          "pre_call_questions": ["Who is sponsor for renewal?","Any blockers from finance?","Treasury appetite and timeline?"],
+          "sources": ["Account:001X","Opportunity:006A","Opportunity:006B","Case:500C","Task:00T1"]
+        }
+    }
+]
+
+def build_prompt(context_json: str) -> str:
+    # Guidance prompt for instruction-tuned models like FLAN-T5
+    parts = []
+    parts.append("You are a banking relationship assistant. Summarize pre-call context for the given Account.")
+    parts.append("Be concise and factual. If data is missing, say 'Unknown' briefly. Output STRICT JSON only.")
+    parts.append(SCHEMA_TEXT)
+    parts.append("Here is a worked example:")
+    parts.append("Context: " + str(FEW_SHOT[0]["context"]))
+    parts.append("Output JSON: " + str(FEW_SHOT[0]["output"]))
+    parts.append("Now produce output for this Context:")
+    parts.append(context_json)
+    parts.append("Output JSON only:")
+    return "\n\n".join(parts)

requirements.txt

@@ -1,3 +1,10 @@
-altair
-pandas
-streamlit
+
+streamlit==1.37.0
+requests==2.32.3
+pydantic==2.8.2
+orjson==3.10.7
+json5==0.9.25
+# If you later want local models instead of Inference API, uncomment below
+# transformers==4.43.3
+# torch>=2.2.0
+# sentencepiece==0.2.0

runtime.txt

@@ -0,0 +1 @@
+python-3.10.14

salesforce.py

@@ -0,0 +1,41 @@
+
+import requests
+from typing import Dict, Any
+
+def sf_query(instance_url: str, access_token: str, soql: str) -> list[dict]:
+    api = f"{instance_url}/services/data/v61.0/query"
+    res = requests.get(api, params={"q": soql},
+                       headers={"Authorization": f"Bearer {access_token}"})
+    res.raise_for_status()
+    return res.json().get("records", [])
+
+def get_context(instance_url: str, access_token: str, account_id: str,
+                limit_opps: int = 5, limit_cases: int = 5, limit_tasks: int = 10) -> Dict[str, Any]:
+    acct = sf_query(instance_url, access_token,
+        f"SELECT Id, Name, Industry, Type, Rating, AnnualRevenue, Owner.Alias, BillingCountry "
+        f"FROM Account WHERE Id='{account_id}' LIMIT 1")
+    opps = sf_query(instance_url, access_token,
+        f"SELECT Id, Name, StageName, Amount, CloseDate, NextStep, Owner.Alias "
+        f"FROM Opportunity WHERE AccountId='{account_id}' ORDER BY LastModifiedDate DESC LIMIT {limit_opps}")
+    cases = sf_query(instance_url, access_token,
+        f"SELECT Id, CaseNumber, Subject, Status, Priority, Origin, LastModifiedDate "
+        f"FROM Case WHERE AccountId='{account_id}' ORDER BY LastModifiedDate DESC LIMIT {limit_cases}")
+    tasks = sf_query(instance_url, access_token,
+        f"SELECT Id, Subject, ActivityDate, Status, Owner.Alias "
+        f"FROM Task WHERE WhatId='{account_id}' OR AccountId='{account_id}' "
+        f"ORDER BY ActivityDate DESC LIMIT {limit_tasks}")
+    return {
+        "account": acct[0] if acct else {},
+        "opportunities": opps,
+        "cases": cases,
+        "activities": tasks
+    }
+
+def create_note(instance_url: str, access_token: str, account_id: str, title: str, body: str) -> str:
+    payload = {"Title": title, "Body": body, "ParentId": account_id}
+    res = requests.post(f"{instance_url}/services/data/v61.0/sobjects/Note",
+                        headers={"Authorization": f"Bearer {access_token}",
+                                 "Content-Type": "application/json"},
+                        json=payload)
+    res.raise_for_status()
+    return res.json().get("id")

schema.py

@@ -0,0 +1,49 @@
+
+from pydantic import BaseModel, Field, ValidationError
+from typing import List, Optional
+
+class TimelineItem(BaseModel):
+    date: str = Field(..., description="YYYY-MM-DD")
+    event: str
+    source: str
+
+class OpportunityItem(BaseModel):
+    name: str
+    stage: str
+    amount: float
+    closeDate: Optional[str] = None
+    nextStep: Optional[str] = None
+
+class PreCallSummary(BaseModel):
+    client_overview: str
+    relationship_timeline: List[TimelineItem]
+    open_items: List[str]
+    opportunities: List[OpportunityItem]
+    risks_flags: List[str]
+    recommendations: List[str]
+    pre_call_questions: List[str]
+    sources: List[str]
+
+def ensure_schema(d: dict) -> PreCallSummary:
+    """Validate and coerce a dict to PreCallSummary, filling missing keys with safe defaults."""
+    defaults = {
+        "client_overview": "",
+        "relationship_timeline": [],
+        "open_items": [],
+        "opportunities": [],
+        "risks_flags": [],
+        "recommendations": [],
+        "pre_call_questions": [],
+        "sources": []
+    }
+    merged = {**defaults, **(d or {})}
+    try:
+        return PreCallSummary.model_validate(merged)
+    except ValidationError as e:
+        # Best-effort repair for common issues
+        for fld in ["relationship_timeline", "open_items", "opportunities", "risks_flags", "recommendations", "pre_call_questions", "sources"]:
+            if merged.get(fld) is None:
+                merged[fld] = []
+        if merged.get("client_overview") is None:
+            merged["client_overview"] = ""
+        return PreCallSummary.model_validate(merged)