Upload 21 files

app.py

@@ -1,35 +1,38 @@
-import os, json, requests, streamlit as st
+
+import os,json,time,streamlit as st
 from backend.rag_engine import get_embedder,get_chroma,retrieve,seed_index
 from backend.soap_generator import compose_soap
-from utils.constants import DOCS_DIR,CHAT_ENDPOINT
+from backend.pdf_utils import generate_pdf
+from backend.endpoint_client import call_endpoint
+from utils.constants import DOCS_DIR,RETRIEVAL_K_DEFAULT
+from utils.persona import AI_GYNO_PERSONA_V2
+
+st.set_page_config(page_title="MediAssist v14.2 Clean",page_icon="🩺",layout="wide")
 
-st.set_page_config(page_title='MediAssist v13',page_icon='🩺',layout='wide')
 @st.cache_resource
 def emb():return get_embedder()
 @st.cache_resource
 def col():return get_chroma()[1]
 
-def chat(prompt):
-    token=os.getenv('HF_API_TOKEN')
-    if not token:return 'Missing HF_API_TOKEN'
-    r=requests.post(CHAT_ENDPOINT,headers={"Authorization":f"Bearer {token}"},json={"inputs":prompt},timeout=200)
-    d=r.json()
-    if isinstance(d,list) and "generated_text" in d[0]:
-        return d[0]["generated_text"]
-    return str(d)
+st.title("🩺 MediAssist v14.2 — Clean Stable Build")
+
+txt=st.text_area("Patient narrative")
+k=st.slider("Results",1,10,RETRIEVAL_K_DEFAULT)
+ep=st.text_input("Endpoint override")
 
-st.title("🩺 MediAssist v13 — AI Gynae Assistant")
+if st.button("Generate OPD"):
+    items=retrieve(col(),emb(),txt,k)
+    soap=compose_soap(txt,items)
+    st.json(soap)
 
-with st.sidebar:
-    if st.button("Seed Index"):
-        n=seed_index(col(),emb(),DOCS_DIR);st.success(f"Indexed {n} chunks")
+if st.button("AI Chat"):
+    prompt=f"{AI_GYNO_PERSONA_V2}\nPatient:{txt}\nAssistant:"
+    reply,_=call_endpoint(prompt,endpoint=ep or None)
+    st.write(reply)
 
-txt=st.text_area("Patient narrative")
-if st.button("Generate Report"):
-    items=retrieve(col(),emb(),txt,5)
+summ=st.text_area("Doctor summary")
+if st.button("Generate PDF"):
+    items=retrieve(col(),emb(),txt,3)
     soap=compose_soap(txt,items)
-    ctx="\n".join([i["text"] for i in items])
-    prompt=f"Use this context to create a refined clinical report:\n{ctx}\nPatient: {txt}"
-    reply=chat(prompt)
-    st.subheader("AI Draft Report");st.write(reply)
-    st.subheader("SOAP");st.json(soap)
+    generate_pdf("report.pdf","MediAssist Report",soap,summ)
+    st.download_button("Download PDF",open("report.pdf","rb"),file_name="report.pdf",mime="application/pdf")

backend/endpoint_client.py

@@ -0,0 +1,26 @@
+
+import os,time,requests,json
+from utils.constants import CHAT_ENDPOINT,MAX_RETRIES_DEFAULT,RETRY_BACKOFF_SECONDS_DEFAULT,REQUEST_TIMEOUT_SECONDS_DEFAULT
+
+def call_endpoint(prompt,endpoint=None,token=None,max_retries=None,backoff=None,timeout=None,logs=None):
+    url=endpoint or os.getenv("CHAT_ENDPOINT") or CHAT_ENDPOINT
+    tok=token or os.getenv("HF_API_TOKEN")
+    if not tok:return "❌ Missing HF_API_TOKEN.",{}
+    mr=max_retries or MAX_RETRIES_DEFAULT
+    bf=backoff or RETRY_BACKOFF_SECONDS_DEFAULT
+    to=timeout or REQUEST_TIMEOUT_SECONDS_DEFAULT
+
+    h={"Authorization":f"Bearer {tok}","Content-Type":"application/json"}
+    for a in range(1,mr+1):
+        try:
+            r=requests.post(url,headers=h,json={"inputs":prompt},timeout=to)
+            try:data=r.json()
+            except:return "⚠️ Non-JSON:\n"+r.text,{}
+            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"],{}
+            return "⚠️ Unexpected:"+json.dumps(data)[:500],{}
+        except:
+            time.sleep(bf*a)
+    return "❌ Endpoint unavailable",{}

backend/pdf_utils.py

@@ -0,0 +1,14 @@
+
+from reportlab.pdfgen import canvas
+from reportlab.lib.pagesizes import A4
+from reportlab.lib.units import mm
+
+def generate_pdf(path,title,soap,summary):
+    c=canvas.Canvas(path,pagesize=A4)
+    x,y=20*mm,270*mm
+    c.setFont("Helvetica-Bold",16);c.drawString(x,y,title);y-=15
+    c.setFont("Helvetica",10)
+    for k,v in soap.items():
+        c.drawString(x,y,f"{k}: {v}");y-=10
+    c.drawString(x,y,f"Doctor Summary: {summary}");y-=10
+    c.save()

backend/rag_engine.py

@@ -1,32 +1,45 @@
-import os,glob,chromadb
-from sentence_transformers import SentenceTransformer,models
+
+import os,glob
+import chromadb
+from sentence_transformers import SentenceTransformer, models
 from langchain_text_splitters import RecursiveCharacterTextSplitter
 from utils.constants import CHROMA_DIR,DOCS_DIR,COLLECTION,EMB_MODEL_NAME
+from utils.helpers import to_safe_items
+
 def get_embedder():
-    w=models.Transformer(EMB_MODEL_NAME);p=models.Pooling(w.get_word_embedding_dimension())
+    w=models.Transformer(EMB_MODEL_NAME)
+    p=models.Pooling(w.get_word_embedding_dimension())
     return SentenceTransformer(modules=[w,p])
+
 def get_chroma():
     c=chromadb.PersistentClient(path=CHROMA_DIR)
-    return c,c.get_or_create_collection(COLLECTION,metadata={"hnsw:space":"cosine"})
-def embed(m,txts):return m.encode(txts,convert_to_numpy=True).tolist()
+    return c,c.get_or_create_collection(COLLECTION)
+
+def embed(m,t):return m.encode(t,convert_to_numpy=True).tolist()
+
 def seed_index(col,m,folder):
-    sp=RecursiveCharacterTextSplitter(chunk_size=1000,chunk_overlap=150)
-    paths=glob.glob(folder+'/*.txt')
+    spl=RecursiveCharacterTextSplitter(chunk_size=1000,chunk_overlap=150)
+    paths=glob.glob(folder+'/**/*.txt',recursive=True)
     ids,docs,meta=[],[],[]
     for p in paths:
-        t=os.path.basename(p).replace('.txt','')
-        with open(p) as f:tx=f.read()
-        for i,ch in enumerate(sp.split_text(tx)):
-            ids.append(f"{t}-{i}");docs.append(ch);meta.append({"title":t,"source":p})
+        try:txt=open(p).read()
+        except:continue
+        title=os.path.basename(p).replace('.txt','')
+        chunks=spl.split_text(txt)
+        for i,ch in enumerate(chunks):
+            ids.append(f"{title}-{i}")
+            docs.append(ch)
+            meta.append({"title":title,"source":p})
+    if not docs:return 0
     em=embed(m,docs)
     try:col.add(ids=ids,documents=docs,metadatas=meta,embeddings=em)
-    except:col.delete(ids=ids);col.add(ids=ids,documents=docs,metadatas=meta,embeddings=em)
+    except:
+        try:col.delete(ids=ids)
+        except:pass
+        col.add(ids=ids,documents=docs,metadatas=meta,embeddings=em)
     return len(docs)
+
 def retrieve(col,m,q,k):
     em=embed(m,[q])[0]
     r=col.query(query_embeddings=[em],n_results=k,include=["documents","metadatas"])
-    out=[]
-    if r.get("ids"):
-        for i in range(len(r["ids"][0])):
-            out.append({"text":r["documents"][0][i],"title":r["metadatas"][0][i]["title"],"source":r["metadatas"][0][i]["source"]})
-    return out
+    return to_safe_items(r)

backend/soap_generator.py

@@ -1,7 +1,11 @@
+
 def compose_soap(n,items):
-    a=[];p=["Follow-up in 1 week","Safety net advice"]
+    a=[];p=["Follow-up in 3–7 days.","Safety-net instructions."]
     t=n.lower()
-    if "bleed" in t:a.append("Abnormal uterine bleeding")
-    if "pain" in t:a.append("Pelvic pain evaluation")
-    if not a:a.append("General gynae evaluation")
-    return {"subjective":n,"assessment":a,"plan":p,"citations":[i["title"] for i in items]}
+    if "pain" in t:a.append("Pain evaluation")
+    if "bleed" in t:a.append("Bleeding assessment")
+    if not a:a.append("General evaluation")
+    return {
+        "subjective":n,"objective":"Vitals stable","assessment":a,"plan":p,
+        "citations":[i["title"] for i in items]
+    }

data/guidelines/internal_med/0_PCOS.txt

@@ -0,0 +1 @@
+PCOS guideline for internal_med

data/guidelines/internal_med/1_AUB.txt

@@ -0,0 +1 @@
+AUB guideline for internal_med

data/guidelines/internal_med/2_Chest Pain.txt

@@ -0,0 +1 @@
+Chest Pain guideline for internal_med

data/guidelines/internal_med/3_Knee Pain.txt

@@ -0,0 +1 @@
+Knee Pain guideline for internal_med

data/guidelines/obgyn/0_PCOS.txt

@@ -0,0 +1 @@
+PCOS guideline for obgyn

data/guidelines/obgyn/1_AUB.txt

@@ -0,0 +1 @@
+AUB guideline for obgyn

data/guidelines/obgyn/2_Chest Pain.txt

@@ -0,0 +1 @@
+Chest Pain guideline for obgyn

data/guidelines/obgyn/3_Knee Pain.txt

@@ -0,0 +1 @@
+Knee Pain guideline for obgyn

data/guidelines/orthopedics/0_PCOS.txt

@@ -0,0 +1 @@
+PCOS guideline for orthopedics

data/guidelines/orthopedics/1_AUB.txt

@@ -0,0 +1 @@
+AUB guideline for orthopedics

data/guidelines/orthopedics/2_Chest Pain.txt

@@ -0,0 +1 @@
+Chest Pain guideline for orthopedics

data/guidelines/orthopedics/3_Knee Pain.txt

@@ -0,0 +1 @@
+Knee Pain guideline for orthopedics

requirements.txt

@@ -3,3 +3,4 @@ chromadb
 sentence-transformers
 langchain-text-splitters
 requests
+reportlab

utils/constants.py

@@ -1,5 +1,10 @@
-CHROMA_DIR='./data/chroma'
-DOCS_DIR='./data/guidelines'
-COLLECTION='gynae_guidelines'
-EMB_MODEL_NAME='medicalai/ClinicalBERT'
-CHAT_ENDPOINT='https://router.huggingface.co/models/openai/gpt-oss-120b'
+
+CHROMA_DIR = "./data/chroma"
+DOCS_DIR = "./data/guidelines"
+COLLECTION = "med_guidelines_multispeciality"
+EMB_MODEL_NAME = "medicalai/ClinicalBERT"
+RETRIEVAL_K_DEFAULT = 5
+CHAT_ENDPOINT = "https://api-inference.huggingface.co/models/openai/gpt-oss-120b"
+MAX_RETRIES_DEFAULT = 6
+RETRY_BACKOFF_SECONDS_DEFAULT = 3
+REQUEST_TIMEOUT_SECONDS_DEFAULT = 60

utils/helpers.py

@@ -0,0 +1,13 @@
+
+from typing import List,Dict
+
+def to_safe_items(res)->List[Dict]:
+    items=[]
+    if res and res.get("ids"):
+        for i in range(len(res["ids"][0])):
+            items.append({
+                "text":res["documents"][0][i],
+                "title":res["metadatas"][0][i].get("title","(untitled)"),
+                "source":res["metadatas"][0][i].get("source",""),
+            })
+    return items

utils/persona.py

@@ -0,0 +1,10 @@
+
+AI_GYNO_PERSONA_V2 = """
+You are AIgyno, a multispeciality clinical assistant for doctors.
+
+- Provide safe, accurate, careful medical reasoning.
+- Use retrieved context.
+- Ask clarifying questions if uncertain.
+- Provide differentials and next steps.
+- Doctor review required for final diagnosis.
+"""