Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -1,5 +1,5 @@
|
|
1 |
# ------------------------------
|
2 |
-
# Imports
|
3 |
# ------------------------------
|
4 |
from langchain_openai import OpenAIEmbeddings
|
5 |
from langchain_community.vectorstores import Chroma
|
@@ -21,7 +21,7 @@ from langchain.tools.retriever import create_retriever_tool
|
|
21 |
from datetime import datetime
|
22 |
|
23 |
# ------------------------------
|
24 |
-
# Data
|
25 |
# ------------------------------
|
26 |
research_texts = [
|
27 |
"Research Report: Results of a New AI Model Improving Image Recognition Accuracy to 98%",
|
@@ -40,7 +40,7 @@ development_texts = [
|
|
40 |
]
|
41 |
|
42 |
# ------------------------------
|
43 |
-
# Configuration
|
44 |
# ------------------------------
|
45 |
class AppConfig:
|
46 |
def __init__(self):
|
@@ -52,49 +52,43 @@ class AppConfig:
|
|
52 |
self.DOCUMENT_OVERLAP = 50
|
53 |
self.SEARCH_K = 5
|
54 |
self.SEARCH_TYPE = "mmr"
|
55 |
-
|
56 |
-
|
57 |
-
def validate_config(self):
|
58 |
if not self.DEEPSEEK_API_KEY:
|
59 |
st.error("""
|
60 |
-
**
|
61 |
-
π DeepSeek API key
|
62 |
-
|
63 |
-
1.
|
64 |
-
2. Add secret:
|
65 |
3. Rebuild Space
|
66 |
""")
|
67 |
st.stop()
|
68 |
|
69 |
-
config = AppConfig()
|
70 |
-
|
71 |
-
# ------------------------------
|
72 |
-
# ChromaDB Manager
|
73 |
-
# ------------------------------
|
74 |
class ChromaManager:
|
75 |
-
def __init__(self,
|
76 |
os.makedirs(config.CHROMA_PATH, exist_ok=True)
|
77 |
self.client = chromadb.PersistentClient(path=config.CHROMA_PATH)
|
78 |
self.embeddings = OpenAIEmbeddings(model="text-embedding-3-large")
|
79 |
|
80 |
-
self.research_collection = self.
|
81 |
-
|
82 |
"research_collection",
|
83 |
-
{"category": "research"
|
84 |
)
|
85 |
-
self.dev_collection = self.
|
86 |
-
|
87 |
"development_collection",
|
88 |
-
{"category": "development"
|
89 |
)
|
90 |
|
91 |
-
def
|
92 |
-
|
93 |
-
chunk_size=
|
94 |
-
chunk_overlap=
|
95 |
-
separators=["\n\n", "\n", "γ"
|
96 |
)
|
97 |
-
docs =
|
98 |
return Chroma.from_documents(
|
99 |
documents=docs,
|
100 |
embedding=self.embeddings,
|
@@ -103,383 +97,208 @@ class ChromaManager:
|
|
103 |
collection_metadata=metadata
|
104 |
)
|
105 |
|
106 |
-
# Initialize Chroma with data
|
107 |
-
chroma_manager = ChromaManager(research_texts, development_texts)
|
108 |
-
|
109 |
# ------------------------------
|
110 |
# Document Processing
|
111 |
# ------------------------------
|
112 |
class DocumentProcessor:
|
113 |
@staticmethod
|
114 |
-
def
|
115 |
seen = set()
|
116 |
-
|
117 |
-
|
118 |
-
|
119 |
-
if content_hash not in seen:
|
120 |
-
unique_docs.append(doc)
|
121 |
-
seen.add(content_hash)
|
122 |
-
return unique_docs
|
123 |
|
124 |
@staticmethod
|
125 |
-
def
|
126 |
-
key_points = []
|
127 |
categories = {
|
128 |
-
"quantum": ["quantum", "
|
129 |
-
"vision": ["image", "recognition"
|
130 |
-
"nlp": ["transformer", "language"
|
131 |
}
|
132 |
-
|
133 |
-
|
134 |
-
|
135 |
-
|
136 |
-
|
137 |
-
|
138 |
-
|
139 |
-
|
140 |
-
|
141 |
-
|
142 |
-
return "\n".join(list(set(key_points)))
|
143 |
|
144 |
# ------------------------------
|
145 |
-
#
|
146 |
# ------------------------------
|
147 |
-
class
|
148 |
-
def __init__(self):
|
149 |
-
self.
|
150 |
-
|
151 |
-
"doc_counts": [],
|
152 |
-
"error_count": 0
|
153 |
-
}
|
154 |
-
|
155 |
-
def api_request_with_retry(self, endpoint: str, payload: Dict) -> Dict:
|
156 |
-
headers = {
|
157 |
-
"Authorization": f"Bearer {config.DEEPSEEK_API_KEY}",
|
158 |
-
"Content-Type": "application/json"
|
159 |
-
}
|
160 |
|
161 |
-
|
162 |
-
|
163 |
-
|
164 |
-
|
165 |
-
|
166 |
-
|
167 |
-
|
168 |
-
|
169 |
-
|
170 |
-
|
171 |
-
|
172 |
-
|
173 |
-
|
174 |
-
|
175 |
-
|
176 |
-
|
177 |
-
raise
|
178 |
-
raise Exception(f"API request failed after {config.MAX_RETRIES} attempts")
|
179 |
|
180 |
-
#
|
181 |
-
|
182 |
-
|
183 |
-
|
184 |
-
|
185 |
-
|
186 |
-
def agent(state: AgentState):
|
187 |
-
print("---CALL AGENT---")
|
188 |
-
messages = state["messages"]
|
189 |
-
user_message = messages[0].content if not isinstance(messages[0], tuple) else messages[0][1]
|
190 |
-
|
191 |
-
prompt = f"""Given this user question: "{user_message}"
|
192 |
-
If about research/academic topics, respond EXACTLY:
|
193 |
-
SEARCH_RESEARCH: <search terms>
|
194 |
-
If about development status, respond EXACTLY:
|
195 |
-
SEARCH_DEV: <search terms>
|
196 |
-
Otherwise, answer directly."""
|
197 |
-
|
198 |
-
headers = {
|
199 |
-
"Accept": "application/json",
|
200 |
-
"Authorization": f"Bearer {config.DEEPSEEK_API_KEY}",
|
201 |
-
"Content-Type": "application/json"
|
202 |
-
}
|
203 |
-
|
204 |
-
data = {
|
205 |
-
"model": "deepseek-chat",
|
206 |
-
"messages": [{"role": "user", "content": prompt}],
|
207 |
-
"temperature": 0.7,
|
208 |
-
"max_tokens": 1024
|
209 |
-
}
|
210 |
-
|
211 |
-
try:
|
212 |
-
response = requests.post(
|
213 |
-
"https://api.deepseek.com/v1/chat/completions",
|
214 |
-
headers=headers,
|
215 |
-
json=data,
|
216 |
-
verify=False,
|
217 |
-
timeout=30
|
218 |
)
|
219 |
-
|
220 |
-
|
221 |
-
|
222 |
-
|
223 |
-
|
224 |
-
|
225 |
-
|
226 |
-
|
227 |
-
elif "SEARCH_DEV:" in response_text:
|
228 |
-
query = response_text.split("SEARCH_DEV:")[1].strip()
|
229 |
-
results = chroma_manager.dev_collection.as_retriever().invoke(query)
|
230 |
-
return {"messages": [AIMessage(content=f'Action: development_db_tool\n{{"query": "{query}"}}\n\nResults: {str(results)}')]}
|
231 |
|
232 |
-
|
233 |
-
except Exception as e:
|
234 |
-
error_msg = f"API Error: {str(e)}"
|
235 |
-
if "Insufficient Balance" in str(e):
|
236 |
-
error_msg += "\n\nPlease check your DeepSeek API account balance."
|
237 |
-
return {"messages": [AIMessage(content=error_msg)]}
|
238 |
|
239 |
-
|
240 |
-
|
241 |
-
last_message = messages[-1]
|
242 |
-
return "generate" if "Results: [Document" in last_message.content else "rewrite"
|
243 |
|
244 |
-
def
|
245 |
-
|
246 |
-
|
247 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
248 |
|
249 |
-
|
250 |
-
|
251 |
-
|
252 |
-
|
253 |
-
|
254 |
-
processed_info = DocumentProcessor.extract_key_points(
|
255 |
-
DocumentProcessor.deduplicate_documents(docs_content)
|
256 |
-
)
|
257 |
-
|
258 |
-
prompt = f"""Generate structured research summary:
|
259 |
-
Key Information:
|
260 |
-
{processed_info}
|
261 |
-
Include:
|
262 |
-
1. Section headings
|
263 |
-
2. Bullet points
|
264 |
-
3. Significance
|
265 |
-
4. Applications"""
|
266 |
-
|
267 |
-
try:
|
268 |
response = requests.post(
|
269 |
"https://api.deepseek.com/v1/chat/completions",
|
270 |
-
headers={
|
271 |
-
"Authorization": f"Bearer {config.DEEPSEEK_API_KEY}",
|
272 |
-
"Content-Type": "application/json"
|
273 |
-
},
|
274 |
json={
|
275 |
"model": "deepseek-chat",
|
276 |
-
"messages": [{
|
277 |
-
|
278 |
-
|
279 |
-
|
280 |
-
|
281 |
-
)
|
282 |
-
|
283 |
-
return {"messages": [AIMessage(content=response
|
284 |
-
except Exception as e:
|
285 |
-
return {"messages": [AIMessage(content=f"Generation Error: {str(e)}")]}
|
286 |
|
287 |
-
def rewrite(state: AgentState):
|
288 |
-
|
289 |
-
original_question = messages[0].content
|
290 |
-
|
291 |
-
try:
|
292 |
response = requests.post(
|
293 |
"https://api.deepseek.com/v1/chat/completions",
|
294 |
-
headers={
|
295 |
-
"Authorization": f"Bearer {config.DEEPSEEK_API_KEY}",
|
296 |
-
"Content-Type": "application/json"
|
297 |
-
},
|
298 |
json={
|
299 |
"model": "deepseek-chat",
|
300 |
"messages": [{
|
301 |
-
"role": "user",
|
302 |
-
"content": f"
|
303 |
-
}]
|
304 |
-
|
305 |
-
|
306 |
-
|
307 |
-
timeout=30
|
308 |
-
)
|
309 |
-
response.raise_for_status()
|
310 |
-
return {"messages": [AIMessage(content=response.json()['choices'][0]['message']['content'])]}
|
311 |
-
except Exception as e:
|
312 |
-
return {"messages": [AIMessage(content=f"Rewrite Error: {str(e)}")]}
|
313 |
|
314 |
-
|
|
|
315 |
|
316 |
-
def
|
317 |
-
|
318 |
-
return "tools" if tools_pattern.match(content) else END
|
319 |
|
320 |
# ------------------------------
|
321 |
-
#
|
322 |
# ------------------------------
|
323 |
-
|
324 |
-
|
325 |
-
|
326 |
-
|
327 |
-
|
328 |
-
|
329 |
-
|
330 |
-
|
331 |
-
|
332 |
-
|
333 |
-
"development_db_tool",
|
334 |
-
"Search development database"
|
335 |
-
)
|
336 |
-
]))
|
337 |
-
workflow.add_node("rewrite", rewrite)
|
338 |
-
workflow.add_node("generate", generate)
|
339 |
-
|
340 |
-
workflow.set_entry_point("agent")
|
341 |
-
workflow.add_conditional_edges("agent", custom_tools_condition, {"tools": "retrieve", END: END})
|
342 |
-
workflow.add_conditional_edges("retrieve", simple_grade_documents, {"generate": "generate", "rewrite": "rewrite"})
|
343 |
-
workflow.add_edge("generate", END)
|
344 |
-
workflow.add_edge("rewrite", "agent")
|
345 |
-
|
346 |
-
app = workflow.compile()
|
347 |
-
|
348 |
-
# ------------------------------
|
349 |
-
# Streamlit UI
|
350 |
-
# ------------------------------
|
351 |
-
class UITheme:
|
352 |
-
primary_color = "#2E86C1"
|
353 |
-
secondary_color = "#28B463"
|
354 |
-
background_color = "#1A1A1A"
|
355 |
-
text_color = "#EAECEE"
|
356 |
-
|
357 |
-
@classmethod
|
358 |
-
def apply(cls):
|
359 |
-
st.markdown(f"""
|
360 |
-
<style>
|
361 |
-
.stApp {{ background-color: {cls.background_color}; color: {cls.text_color}; }}
|
362 |
-
.stTextArea textarea {{
|
363 |
-
background-color: #2D2D2D !important;
|
364 |
-
color: {cls.text_color} !important;
|
365 |
-
border: 1px solid {cls.primary_color};
|
366 |
-
}}
|
367 |
-
.stButton > button {{
|
368 |
-
background-color: {cls.primary_color};
|
369 |
-
color: white;
|
370 |
-
border: none;
|
371 |
-
padding: 12px 28px;
|
372 |
-
border-radius: 6px;
|
373 |
-
transition: all 0.3s ease;
|
374 |
-
font-weight: 500;
|
375 |
-
}}
|
376 |
-
.stButton > button:hover {{
|
377 |
-
background-color: {cls.secondary_color};
|
378 |
-
transform: translateY(-1px);
|
379 |
-
box-shadow: 0 4px 12px rgba(0,0,0,0.2);
|
380 |
-
}}
|
381 |
-
.data-box {{
|
382 |
-
background-color: #2D2D2D;
|
383 |
-
border-left: 4px solid {cls.primary_color};
|
384 |
-
padding: 18px;
|
385 |
-
margin: 14px 0;
|
386 |
-
border-radius: 8px;
|
387 |
-
box-shadow: 0 2px 8px rgba(0,0,0,0.15);
|
388 |
-
}}
|
389 |
-
</style>
|
390 |
-
""", unsafe_allow_html=True)
|
391 |
|
392 |
def main():
|
393 |
-
|
394 |
-
|
395 |
-
st.set_page_config(
|
396 |
-
page_title="AI Research Assistant Pro",
|
397 |
-
layout="wide",
|
398 |
-
initial_sidebar_state="expanded",
|
399 |
-
menu_items={
|
400 |
-
'Get Help': 'https://example.com/docs',
|
401 |
-
'Report a bug': 'https://example.com/issues',
|
402 |
-
'About': "v2.1 | Enhanced Research Assistant"
|
403 |
-
}
|
404 |
-
)
|
405 |
|
406 |
with st.sidebar:
|
407 |
-
st.header("
|
408 |
-
with st.expander("Research
|
409 |
for text in research_texts:
|
410 |
-
st.markdown(f'<div class="data-box
|
411 |
-
|
412 |
-
with st.expander("Development Database"):
|
413 |
for text in development_texts:
|
414 |
-
st.markdown(f'<div class="data-box
|
415 |
-
|
416 |
-
st.title("
|
417 |
-
st.
|
418 |
-
|
419 |
-
query = st.text_area(
|
420 |
-
"Research Query Input",
|
421 |
-
height=120,
|
422 |
-
placeholder="Enter your research question...",
|
423 |
-
help="Be specific about domains for better results"
|
424 |
-
)
|
425 |
|
426 |
-
|
427 |
-
|
428 |
-
|
429 |
-
|
430 |
-
|
431 |
-
|
|
|
432 |
|
433 |
-
|
434 |
-
|
435 |
-
start_time = time.time()
|
436 |
-
events = process_question(query, app, {"configurable": {"thread_id": "1"}})
|
437 |
-
|
438 |
-
processed_data = []
|
439 |
-
for event in events:
|
440 |
-
if 'agent' in event:
|
441 |
-
content = event['agent']['messages'][0].content
|
442 |
-
if "Results:" in content:
|
443 |
-
docs = eval(content.split("Results: ")[1])
|
444 |
-
unique_docs = DocumentProcessor.deduplicate_documents(docs)
|
445 |
-
key_points = DocumentProcessor.extract_key_points(unique_docs)
|
446 |
-
processed_data.append(key_points)
|
447 |
-
|
448 |
-
with st.expander("π Retrieved Documents", expanded=False):
|
449 |
-
st.info(f"Found {len(unique_docs)} unique documents")
|
450 |
-
st.write(docs)
|
451 |
-
|
452 |
-
elif 'generate' in event:
|
453 |
-
final_answer = event['generate']['messages'][0].content
|
454 |
-
status.update(label="β
Analysis Complete", state="complete")
|
455 |
-
st.markdown("## π Research Summary")
|
456 |
-
st.markdown(final_answer)
|
457 |
-
|
458 |
-
st.caption(f"β±οΈ Processed in {time.time()-start_time:.2f}s | {len(processed_data)} clusters")
|
459 |
-
|
460 |
-
except Exception as e:
|
461 |
-
status.update(label="β Processing Failed", state="error")
|
462 |
-
st.error(f"**Error:** {str(e)}\n\nCheck API key and network connection")
|
463 |
-
with open("error_log.txt", "a") as f:
|
464 |
-
f.write(f"{datetime.now()} | {str(e)}\n")
|
465 |
-
|
466 |
-
with col2:
|
467 |
-
st.markdown("""
|
468 |
-
## π Usage Guide
|
469 |
-
**1. Query Formulation**
|
470 |
-
- Specify domains (e.g., "quantum NLP")
|
471 |
-
- Include timeframes for recent advances
|
472 |
-
|
473 |
-
**2. Results Interpretation**
|
474 |
-
- Expand sections for source documents
|
475 |
-
- Key points show technical breakthroughs
|
476 |
-
- Summary includes commercial implications
|
477 |
-
|
478 |
-
**3. Advanced Features**
|
479 |
-
- Use keyboard shortcuts for efficiency
|
480 |
-
- Click documents for raw context
|
481 |
-
- Export via screenshot/PDF
|
482 |
-
""")
|
483 |
|
|
|
|
|
|
|
484 |
if __name__ == "__main__":
|
485 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
# ------------------------------
|
2 |
+
# Imports
|
3 |
# ------------------------------
|
4 |
from langchain_openai import OpenAIEmbeddings
|
5 |
from langchain_community.vectorstores import Chroma
|
|
|
21 |
from datetime import datetime
|
22 |
|
23 |
# ------------------------------
|
24 |
+
# Data
|
25 |
# ------------------------------
|
26 |
research_texts = [
|
27 |
"Research Report: Results of a New AI Model Improving Image Recognition Accuracy to 98%",
|
|
|
40 |
]
|
41 |
|
42 |
# ------------------------------
|
43 |
+
# Configuration
|
44 |
# ------------------------------
|
45 |
class AppConfig:
|
46 |
def __init__(self):
|
|
|
52 |
self.DOCUMENT_OVERLAP = 50
|
53 |
self.SEARCH_K = 5
|
54 |
self.SEARCH_TYPE = "mmr"
|
55 |
+
|
56 |
+
def validate(self):
|
|
|
57 |
if not self.DEEPSEEK_API_KEY:
|
58 |
st.error("""
|
59 |
+
**Configuration Error**
|
60 |
+
π Missing DeepSeek API key.
|
61 |
+
Configure through Hugging Face Space secrets:
|
62 |
+
1. Space Settings β Repository secrets
|
63 |
+
2. Add secret: DEEPSEEK_API_KEY=your_key
|
64 |
3. Rebuild Space
|
65 |
""")
|
66 |
st.stop()
|
67 |
|
|
|
|
|
|
|
|
|
|
|
68 |
class ChromaManager:
|
69 |
+
def __init__(self, config: AppConfig):
|
70 |
os.makedirs(config.CHROMA_PATH, exist_ok=True)
|
71 |
self.client = chromadb.PersistentClient(path=config.CHROMA_PATH)
|
72 |
self.embeddings = OpenAIEmbeddings(model="text-embedding-3-large")
|
73 |
|
74 |
+
self.research_collection = self._create_collection(
|
75 |
+
research_texts,
|
76 |
"research_collection",
|
77 |
+
{"category": "research"}
|
78 |
)
|
79 |
+
self.dev_collection = self._create_collection(
|
80 |
+
development_texts,
|
81 |
"development_collection",
|
82 |
+
{"category": "development"}
|
83 |
)
|
84 |
|
85 |
+
def _create_collection(self, documents: List[str], name: str, metadata: dict) -> Chroma:
|
86 |
+
splitter = RecursiveCharacterTextSplitter(
|
87 |
+
chunk_size=300,
|
88 |
+
chunk_overlap=50,
|
89 |
+
separators=["\n\n", "\n", "γ"]
|
90 |
)
|
91 |
+
docs = splitter.create_documents(documents)
|
92 |
return Chroma.from_documents(
|
93 |
documents=docs,
|
94 |
embedding=self.embeddings,
|
|
|
97 |
collection_metadata=metadata
|
98 |
)
|
99 |
|
|
|
|
|
|
|
100 |
# ------------------------------
|
101 |
# Document Processing
|
102 |
# ------------------------------
|
103 |
class DocumentProcessor:
|
104 |
@staticmethod
|
105 |
+
def deduplicate(docs: List[Any]) -> List[Any]:
|
106 |
seen = set()
|
107 |
+
return [doc for doc in docs
|
108 |
+
if not (hashlib.md5(doc.page_content.encode()).hexdigest() in seen
|
109 |
+
or seen.add(hashlib.md5(doc.page_content.encode()).hexdigest()))]
|
|
|
|
|
|
|
|
|
110 |
|
111 |
@staticmethod
|
112 |
+
def extract_keypoints(docs: List[Any]) -> str:
|
|
|
113 |
categories = {
|
114 |
+
"quantum": ["quantum", "qubit"],
|
115 |
+
"vision": ["image", "recognition"],
|
116 |
+
"nlp": ["transformer", "language"]
|
117 |
}
|
118 |
+
return "\n".join(sorted({
|
119 |
+
"- " + {
|
120 |
+
"quantum": "Quantum computing breakthroughs",
|
121 |
+
"vision": "Computer vision advancements",
|
122 |
+
"nlp": "NLP architecture innovations"
|
123 |
+
}[cat]
|
124 |
+
for doc in docs
|
125 |
+
for cat, kw in categories.items()
|
126 |
+
if any(k in doc.page_content.lower() for k in kw)
|
127 |
+
}))
|
|
|
128 |
|
129 |
# ------------------------------
|
130 |
+
# Workflow
|
131 |
# ------------------------------
|
132 |
+
class AgentWorkflow:
|
133 |
+
def __init__(self, chroma: ChromaManager):
|
134 |
+
self.chroma = chroma
|
135 |
+
self.workflow = StateGraph(AgentState)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
136 |
|
137 |
+
# Define nodes
|
138 |
+
self.workflow.add_node("agent", self.agent)
|
139 |
+
self.workflow.add_node("retrieve", ToolNode([
|
140 |
+
create_retriever_tool(
|
141 |
+
chroma.research_collection.as_retriever(),
|
142 |
+
"research_tool",
|
143 |
+
"Search research documents"
|
144 |
+
),
|
145 |
+
create_retriever_tool(
|
146 |
+
chroma.dev_collection.as_retriever(),
|
147 |
+
"dev_tool",
|
148 |
+
"Search development updates"
|
149 |
+
)
|
150 |
+
]))
|
151 |
+
self.workflow.add_node("generate", self.generate)
|
152 |
+
self.workflow.add_node("rewrite", self.rewrite)
|
|
|
|
|
153 |
|
154 |
+
# Define edges
|
155 |
+
self.workflow.set_entry_point("agent")
|
156 |
+
self.workflow.add_conditional_edges(
|
157 |
+
"agent",
|
158 |
+
self._tools_condition,
|
159 |
+
{"retrieve": "retrieve", "end": END}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
160 |
)
|
161 |
+
self.workflow.add_conditional_edges(
|
162 |
+
"retrieve",
|
163 |
+
self._grade_documents,
|
164 |
+
{"generate": "generate", "rewrite": "rewrite"}
|
165 |
+
)
|
166 |
+
self.workflow.add_edge("generate", END)
|
167 |
+
self.workflow.add_edge("rewrite", "agent")
|
|
|
|
|
|
|
|
|
|
|
168 |
|
169 |
+
self.app = self.workflow.compile()
|
|
|
|
|
|
|
|
|
|
|
170 |
|
171 |
+
class AgentState(TypedDict):
|
172 |
+
messages: Annotated[Sequence[AIMessage | HumanMessage | ToolMessage], add_messages]
|
|
|
|
|
173 |
|
174 |
+
def agent(self, state: AgentState):
|
175 |
+
try:
|
176 |
+
messages = state["messages"]
|
177 |
+
query = messages[-1].content if isinstance(messages[-1], HumanMessage) else messages[-1]['content']
|
178 |
+
|
179 |
+
response = requests.post(
|
180 |
+
"https://api.deepseek.com/v1/chat/completions",
|
181 |
+
headers={"Authorization": f"Bearer {config.DEEPSEEK_API_KEY}"},
|
182 |
+
json={
|
183 |
+
"model": "deepseek-chat",
|
184 |
+
"messages": [{
|
185 |
+
"role": "user",
|
186 |
+
"content": f"""Analyze this query: "{query}"
|
187 |
+
Respond EXACTLY as:
|
188 |
+
- SEARCH_RESEARCH: <terms> (for research topics)
|
189 |
+
- SEARCH_DEV: <terms> (for development updates)
|
190 |
+
- DIRECT: <answer> (otherwise)"""
|
191 |
+
}]
|
192 |
+
}
|
193 |
+
).json()
|
194 |
+
|
195 |
+
content = response['choices'][0]['message']['content']
|
196 |
+
if "SEARCH_RESEARCH:" in content:
|
197 |
+
terms = content.split("SEARCH_RESEARCH:")[1].strip()
|
198 |
+
results = self.chroma.research_collection.similarity_search(terms)
|
199 |
+
return {"messages": [AIMessage(content=f"Research Results: {str(results)}")]}
|
200 |
+
elif "SEARCH_DEV:" in content:
|
201 |
+
terms = content.split("SEARCH_DEV:")[1].strip()
|
202 |
+
results = self.chroma.dev_collection.similarity_search(terms)
|
203 |
+
return {"messages": [AIMessage(content=f"Development Results: {str(results)}")]}
|
204 |
+
return {"messages": [AIMessage(content=content)]}
|
205 |
+
|
206 |
+
except Exception as e:
|
207 |
+
return {"messages": [AIMessage(content=f"Error: {str(e)}")]}
|
208 |
|
209 |
+
def generate(self, state: AgentState):
|
210 |
+
docs = eval(state["messages"][-1].content.split("Results: ")[1])
|
211 |
+
processed = "\n".join([d.page_content[:200] for d in DocumentProcessor.deduplicate(docs)])
|
212 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
213 |
response = requests.post(
|
214 |
"https://api.deepseek.com/v1/chat/completions",
|
215 |
+
headers={"Authorization": f"Bearer {config.DEEPSEEK_API_KEY}"},
|
|
|
|
|
|
|
216 |
json={
|
217 |
"model": "deepseek-chat",
|
218 |
+
"messages": [{
|
219 |
+
"role": "user",
|
220 |
+
"content": f"Summarize these findings:\n{processed}"
|
221 |
+
}]
|
222 |
+
}
|
223 |
+
).json()
|
224 |
+
|
225 |
+
return {"messages": [AIMessage(content=response['choices'][0]['message']['content'])]}
|
|
|
|
|
226 |
|
227 |
+
def rewrite(self, state: AgentState):
|
228 |
+
original = state["messages"][0].content
|
|
|
|
|
|
|
229 |
response = requests.post(
|
230 |
"https://api.deepseek.com/v1/chat/completions",
|
231 |
+
headers={"Authorization": f"Bearer {config.DEEPSEEK_API_KEY}"},
|
|
|
|
|
|
|
232 |
json={
|
233 |
"model": "deepseek-chat",
|
234 |
"messages": [{
|
235 |
+
"role": "user",
|
236 |
+
"content": f"Rephrase this query: {original}"
|
237 |
+
}]
|
238 |
+
}
|
239 |
+
).json()
|
240 |
+
return {"messages": [AIMessage(content=response['choices'][0]['message']['content'])]}
|
|
|
|
|
|
|
|
|
|
|
|
|
241 |
|
242 |
+
def _tools_condition(self, state: AgentState):
|
243 |
+
return "retrieve" if "Results:" in state["messages"][-1].content else "end"
|
244 |
|
245 |
+
def _grade_documents(self, state: AgentState):
|
246 |
+
return "generate" if len(eval(state["messages"][-1].content.split("Results: ")[1])) > 0 else "rewrite"
|
|
|
247 |
|
248 |
# ------------------------------
|
249 |
+
# Streamlit App
|
250 |
# ------------------------------
|
251 |
+
def apply_theme():
|
252 |
+
st.markdown("""
|
253 |
+
<style>
|
254 |
+
.stApp { background: #1a1a1a; color: white; }
|
255 |
+
.stTextArea textarea { background: #2d2d2d !important; color: white !important; }
|
256 |
+
.stButton>button { background: #2E86C1; transition: 0.3s; }
|
257 |
+
.stButton>button:hover { background: #1B4F72; transform: scale(1.02); }
|
258 |
+
.data-box { background: #2d2d2d; border-left: 4px solid #2E86C1; padding: 15px; margin: 10px 0; }
|
259 |
+
</style>
|
260 |
+
""", unsafe_allow_html=True)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
261 |
|
262 |
def main():
|
263 |
+
apply_theme()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
264 |
|
265 |
with st.sidebar:
|
266 |
+
st.header("π Databases")
|
267 |
+
with st.expander("Research", expanded=True):
|
268 |
for text in research_texts:
|
269 |
+
st.markdown(f'<div class="data-box">{text}</div>', unsafe_allow_html=True)
|
270 |
+
with st.expander("Development"):
|
|
|
271 |
for text in development_texts:
|
272 |
+
st.markdown(f'<div class="data-box">{text}</div>', unsafe_allow_html=True)
|
273 |
+
|
274 |
+
st.title("π AI Research Assistant")
|
275 |
+
query = st.text_area("Enter your query:", height=100)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
276 |
|
277 |
+
if st.button("Analyze"):
|
278 |
+
with st.spinner("Processing..."):
|
279 |
+
workflow = AgentWorkflow(chroma_manager)
|
280 |
+
results = workflow.app.invoke({"messages": [HumanMessage(content=query)]})
|
281 |
+
|
282 |
+
with st.expander("Processing Details", expanded=True):
|
283 |
+
st.write("### Raw Results", results)
|
284 |
|
285 |
+
st.success("### Final Answer")
|
286 |
+
st.markdown(results['messages'][-1].content)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
287 |
|
288 |
+
# ------------------------------
|
289 |
+
# Initialization
|
290 |
+
# ------------------------------
|
291 |
if __name__ == "__main__":
|
292 |
+
st.set_page_config(
|
293 |
+
page_title="AI Research Assistant",
|
294 |
+
layout="wide",
|
295 |
+
initial_sidebar_state="expanded"
|
296 |
+
)
|
297 |
+
|
298 |
+
try:
|
299 |
+
config = AppConfig()
|
300 |
+
config.validate()
|
301 |
+
chroma_manager = ChromaManager(config)
|
302 |
+
main()
|
303 |
+
except Exception as e:
|
304 |
+
st.error(f"Initialization failed: {str(e)}")
|