Ukubonwa Kwebanga Lokukhiqiza Kwabenzeli be-AI: Ikhodi Encane, Indlela Yokumisa-Yokuqala
kukhula okuyinkimbinkimbi, ukugawulwa kwemithi nokuqapha akuphumeleli. Okudingwa amaqembu empeleni ukubonwa: ikhono lokulandelela izinqumo ze-ejenti, ukuhlola izinga lokuphendula ngokuzenzakalelayo, nokubona ukukhukhuleka ngokuhamba kwesikhathi—ngaphandle kokubhala nokugcina inani elikhulu lokuhlola ngokwezifiso kanye nekhodi ye-telemetry.
Ngakho-ke, amaqembu adinga ukusebenzisa inkundla efanele ukuze abonakale kuyilapho egxile emsebenzini oyinhloko wokwakha nokuthuthukisa i-orchestration yama-ejenti. Futhi hlanganisa uhlelo lwabo lokusebenza endaweni ebonakalayo nge-overhead encane kumakhodi abo okusebenza. Kulesi sihloko, ngizokhombisa ukuthi ungamisa kanjani inkundla yokubuka ye-AI evulekile yomthombo ukuze wenze okulandelayo usebenzisa indlela yekhodi encane:
- LLM-njengomahluleli: Lungiselela abahloli abakhelwe ngaphambilini ukuze bathole izimpendulo Zokulunga, Ukuhambisana, I-Hallucination nokuningi. Bonisa izikolo kuwo wonke ama-run namalogi anemininingwane nezibalo.
- Ukuhlola esikalini: Setha amasethi wedatha ukuze ugcine izimo zokuhlolwa kokuhlehla ukuze ulinganise ukunemba ngokumelene nezimpendulo zeqiniso eliyisisekelo. Thola ngokuqhubekayo i-LLM kanye nokukhukhuleka komenzeli.
- Idatha ye-MELT: Landelela amamethrikhi (ukubambezeleka, ukusetshenziswa kwamathokheni, imodeli yokushayela), imicimbi (amakholi we-API, amakholi we-LLM, ukusetshenziswa kwamathuluzi), amalogi (ukusebenzelana komsebenzisi, ukusebenzisa ithuluzi, ukwenza izinqumo ze-ejenti) nemikhondo enemininingwane – konke ngaphandle kwe-telemetry enemininingwane nekhodi yezinsimbi.
Sizosebenzisa i-Langfuse ukuze sibonakale. Iwumthombo ovulekile kanye ne-framework-agnostic futhi ingasebenza nezinhlaka ze-orchestration ezidumile nabahlinzeki be-LLM.
Uhlelo lokusebenza lwama-ejenti amaningi
Kulo mboniso, nginamathisele ikhodi ye-LangGraph yohlelo lokusebenza Lwesevisi Yekhasimende. Uhlelo lokusebenza lwamukela amathikithi avela kumsebenzisi, luhlukaniselwe Kwezobuchwepheshe, Inkokhiso noma Kokubili kusetshenziswa i-ejenti ye-Triage, bese iluthumela kumenzeli Wosekelo Lokusebenza, umenzeli Wosekelo Lwenkokhelo noma kubo bobabili. Bese i-ejenti yokuphothula ihlanganisa impendulo evela kubo bobabili abasebenzeli ibe yifomethi ehlangene, efundeka kakhudlwana. I-flowchart imi kanje:
Ikhodi inamathiselwe lapha
# --------------------------------------------------
# 0. Load .env
# --------------------------------------------------
from dotenv import load_dotenv
load_dotenv(override=True)
# --------------------------------------------------
# 1. Imports
# --------------------------------------------------
import os
from typing import TypedDict
from langgraph.graph import StateGraph, END
from langchain_openai import AzureChatOpenAI
from langfuse import Langfuse
from langfuse.langchain import CallbackHandler
# --------------------------------------------------
# 2. Langfuse Client (WORKING CONFIG)
# --------------------------------------------------
langfuse = Langfuse(
host="
public_key=os.environ["LANGFUSE_PUBLIC_KEY"] ,
secret_key=os.environ["LANGFUSE_SECRET_KEY"]
)
langfuse_callback = CallbackHandler()
os.environ["LANGGRAPH_TRACING"] = "false"
# --------------------------------------------------
# 3. Azure OpenAI Setup
# --------------------------------------------------
llm = AzureChatOpenAI(
azure_deployment=os.environ["AZURE_OPENAI_DEPLOYMENT_NAME"],
api_version=os.environ.get("AZURE_OPENAI_API_VERSION", "2025-01-01-preview"),
temperature=0.2,
callbacks=[langfuse_callback], # 🔑 enables token usage
)
# --------------------------------------------------
# 4. Shared State
# --------------------------------------------------
class AgentState(TypedDict, total=False):
ticket: str
category: str
technical_response: str
billing_response: str
final_response: str
# --------------------------------------------------
# 5. Agent Definitions
# --------------------------------------------------
def triage_agent(state: dict) -> dict:
with langfuse.start_as_current_observation(
as_type="span",
name="triage_agent",
input={"ticket": state["ticket"]},
) as span:
span.update_trace(name="Customer Service Query - LangGraph Demo")
response = llm.invoke([
{
"role": "system",
"content": (
"Classify the query as one of: "
"Technical, Billing, Both. "
"Respond with only the label."
),
},
{"role": "user", "content": state["ticket"]},
])
raw = response.content.strip().lower()
if "both" in raw:
category = "Both"
elif "technical" in raw:
category = "Technical"
elif "billing" in raw:
category = "Billing"
else:
category = "Technical" # âś… safe fallback
span.update(output={"raw": raw, "category": category})
return {"category": category}
def technical_support_agent(state: dict) -> dict:
with langfuse.start_as_current_observation(
as_type="span",
name="technical_support_agent",
input={
"ticket": state["ticket"],
"category": state.get("category"),
},
) as span:
response = llm.invoke([
{
"role": "system",
"content": (
"You are a technical support specialist. "
"Provide a clear, step-by-step solution."
),
},
{"role": "user", "content": state["ticket"]},
])
answer = response.content
span.update(output={"technical_response": answer})
return {"technical_response": answer}
def billing_support_agent(state: dict) -> dict:
with langfuse.start_as_current_observation(
as_type="span",
name="billing_support_agent",
input={
"ticket": state["ticket"],
"category": state.get("category"),
},
) as span:
response = llm.invoke([
{
"role": "system",
"content": (
"You are a billing support specialist. "
"Answer clearly about payments, invoices, or accounts."
),
},
{"role": "user", "content": state["ticket"]},
])
answer = response.content
span.update(output={"billing_response": answer})
return {"billing_response": answer}
def finalizer_agent(state: dict) -> dict:
with langfuse.start_as_current_observation(
as_type="span",
name="finalizer_agent",
input={
"ticket": state["ticket"],
"technical": state.get("technical_response"),
"billing": state.get("billing_response"),
},
) as span:
parts = [
f"Technical:n{state['technical_response']}"
for k in ["technical_response"]
if state.get(k)
] + [
f"Billing:n{state['billing_response']}"
for k in ["billing_response"]
if state.get(k)
]
if not parts:
final = "Error: No agent responses available."
else:
response = llm.invoke([
{
"role": "system",
"content": (
"Combine the following agent responses into ONE clear, professional, "
"customer-facing answer. Do not mention agents or internal labels. "
f"Answer the user's query: '{state['ticket']}'."
),
},
{"role": "user", "content": "nn".join(parts)},
])
final = response.content
span.update(output={"final_response": final})
return {"final_response": final}
# --------------------------------------------------
# 6. LangGraph Construction
# --------------------------------------------------
builder = StateGraph(AgentState)
builder.add_node("triage", triage_agent)
builder.add_node("technical", technical_support_agent)
builder.add_node("billing", billing_support_agent)
builder.add_node("finalizer", finalizer_agent)
builder.set_entry_point("triage")
# Conditional routing
builder.add_conditional_edges(
"triage",
lambda state: state["category"],
{
"Technical": "technical",
"Billing": "billing",
"Both": "technical",
"__default__": "technical", # âś… never dead-end
},
)
# Sequential resolution
builder.add_conditional_edges(
"technical",
lambda state: state["category"],
{
"Both": "billing", # Proceed to billing if Both
"__default__": "finalizer",
},
)
builder.add_edge("billing", "finalizer")
builder.add_edge("finalizer", END)
graph = builder.compile()
# --------------------------------------------------
# 9. Main
# --------------------------------------------------
if __name__ == "__main__":
print("===============================================")
print(" Conditional Multi-Agent Support System (Ready)")
print("===============================================")
print("Enter 'exit' or 'quit' to stop the program.n")
while True:
# Get user input for the ticket
ticket = input("Enter your support query (ticket): ")
# Check for exit command
if ticket.lower() in ["exit", "quit"]:
print("nExiting the support system. Goodbye!")
break
if not ticket.strip():
print("Please enter a non-empty query.")
continue
try:
# --- Run the graph with the user's ticket ---
result = graph.invoke(
{"ticket": ticket},
config={"callbacks": [langfuse_callback]},
)
# --- Print Results ---
category = result.get('category', 'N/A')
print(f"nâś… Triage Classification: **{category}**")
# Check which agents were executed based on the presence of a response
executed_agents = []
if result.get("technical_response"):
executed_agents.append("Technical")
if result.get("billing_response"):
executed_agents.append("Billing")
print(f"🛠️ Agents Executed: {', '.join(executed_agents) if executed_agents else 'None (Triage Failed)'}")
print("n================ FINAL RESPONSE ================n")
print(result["final_response"])
print("n" + "="*60 + "n")
except Exception as e:
# This is important for debugging: print the exception type and message
print(f"nAn error occurred during processing ({type(e).__name__}): {e}")
print("nPlease try another query.")
print("n" + "="*60 + "n")
Ukucushwa kokubonakala
Ukuze usethe i-Langfuse, hamba futhi usethe i-akhawunti enesigaba Senkokhelo (isigaba sokuzilibazisa esinemikhawulo eningi etholakalayo), bese usetha Iphrojekthi. Kuzilungiselelo zephrojekthi, ungakhiqiza okhiye basesidlangalaleni nabayimfihlo okudingeka banikezwe ekuqaleni kwekhodi. Udinga futhi ukwengeza uxhumano lwe-LLM, oluzosetshenziselwa ukuhlolwa kwe-LLM-as-a-Jaji.
Ukusethwa kwe-LLM-as-a-Jaji
Lona umnyombo wokusethwa kokuhlolwa kokusebenza kwama-ejenti. Lapha ungalungiselela Abahloli abahlukahlukene abakhelwe ngaphambilini kusukela Kulabhulali Yokuhlola ezokora izimpendulo ngemibandela ehlukahlukene efana nokufingqa, Ukulunga, I-Hallucination, Ukugxeka Okuphendulayo njll. Lokhu kufanele kwanele ezimweni eziningi zokusetshenziswa, ngaphandle kwalokho Abahloli Ngokwezifiso nabo bangasethwa. Nakhu ukubuka kwelabhulali yoMhloli:
Khetha umhloli, yisho Ukuhambisana, ofisa ukusisebenzisa. Ungakhetha ukuyisebenzisela ukulandelelwa okusha noma okukhona noma ukugijima kwe-Dataset. Ngaphezu kwalokho, buyekeza umyalo wokuhlola ukuze uqinisekise ukuthi wenelisa umgomo wakho wokuhlola. Okubaluleke kakhulu, umbuzo, isizukulwane kanye nokunye okuguquguqukayo kufanele kufakwe imephu ngendlela efanele emthonjeni (imvamisa, iye Kokufakayo kanye Nokukhiphayo kusuka kumkhondo wohlelo lokusebenza). Okwethu, lokhu kuzoba yidatha yethikithi efakwe umsebenzisi kanye nempendulo ekhiqizwe i-ejenti yokuphothula ngokulandelanayo. Ngaphezu kwalokho, ngokuqalisa kwe-Dataset, ungakwazi ukuqhathanisa izimpendulo ezikhiqiziwe nezimpendulo Zeqiniso Eliphansi ezigcinwe njengemiphumela elindelekile (echazwe ezigabeni ezilandelayo).
Nakhu ukucushwa kwe-'Ukunemba kwe-GT' ukuhlola engikusethele ukuqalisa okusha kwe-Dataset, kanye nokushintshashintsha kwemephu. Ukuhlola kuqala kokwaziswa kokuhlola nakho kuyavezwa. Iningi labahloli lithola amaphuzu phakathi kwebanga elingu-0 kuye ku-1:
Kudemo yesevisi yamakhasimende, ngilungiselele abahloli abangu-3 – Ukuhambisana, Ukufingqa egijima kuwo wonke amathrekhi amasha, futhi Ukunemba kwe-GTesetshenziswa ku-Dataset isebenza kuphela.
Ukusethwa kwedathasethi
Dala idathasethi ezosetshenziswa njengendawo yokugcina icala. Lapha, ungagcina izimo zokuhlola ngombuzo wokufakwayo kanye nempendulo elindelekile efanelekile. Ukuze udale idathasethi, kunezinketho ezi-3: dala irekhodi elilodwa ngesikhathi, layisha i-CSV yemibuzo nezimpendulo ezilindelekile, noma, kalula nje, engeza okokufaka nokuphumayo. ngokuqondile kusuka ekulandeleni kohlelo lokusebenza ezimpendulo zabo zithathwa njengezisezingeni elihle ngochwepheshe abangabantu.
Nali idathasethi engiyidalele idemo. Lokhu kuyingxube yemibuzo yobuchwepheshe, yokukhokha, noma 'Yomibili', futhi ngidale wonke amarekhodi kusuka ekulandeleni kwezicelo:
Yilokho kuphela! Ukucushwa kwenziwa futhi sesilungele ukuqalisa ukubonwa.
Imiphumela Yokubonwa
Ikhasi Lasekhaya le-Langfuse liyideshibhodi yamashadi amaningana awusizo. Ibonisa isibalo sokulandela umkhondo, amaphuzu nama-avareji ngokubuka nje, ukulandelelwa kwesikhathi, ukusetshenziswa kwemodeli nezindleko njll.
Idatha ye-MELT
Idatha yokubonakala ewusizo kakhulu iyatholakala kunketho 'Yokulandela', ebonisa ukubuka okufingqiwe nokunemininingwane yakho konke ukubulawa. Nakhu ukubuka kwedeshibhodi ebonisa isikhathi, igama, okokufaka, okukhiphayo kanye nokubambezeleka okubalulekile namamethrikhi okusebenzisa amathokheni. Qaphela ukuthi kukho konke ukwenziwa kwe-ejenti yohlelo lwethu lokusebenza, kukhona imikhondo yokuhlola emi-2 eyenzelwe Ukufingqa futhi Ukuhambisana abahloli esibamisayo.
Ake sibheke imininingwane yokukodwa kokwenziwa kohlelo lokusebenza Lwesevisi Yekhasimende. Kuphaneli engakwesokunxele, ukugeleza kwe-ejenti kuboniswa kokubili njengesihlahla kanye ne-flowchart. Ibonisa ama-LangGraph nodes (ama-ejenti) kanye nezingcingo ze-LLM kanye nokusetshenziswa kwamathokheni. Ukube abenzeli bethu bebenezingcingo zamathuluzi noma izinyathelo zomuntu, ngabe baboniswe lapha futhi. Qaphela ukuthi amaphuzu okuhlola Ukufingqa futhi Ukuhambisana nazo ziboniswa phezulu, okungu-0.40 no-1 ngokulandelana kwalokhu kugijima. Ukuchofoza kuzo kukhombisa isizathu somphumela kanye nesixhumanisi sokusiyisa kumkhondo womhloli.
Ngakwesokudla, kumenzeli ngamunye, i-LLM kanye nekholi yamathuluzi, singabona Okokufaka kanye nokuphumayo okukhiqizwayo. Isibonelo, lapha sibona ukuthi umbuzo uhlukaniswe ngokuthi 'Kokubili', ngakho-ke eshadini elingakwesokunxele, libonisa kokubili ama-ejenti asekelayo okusebenza nokukhokha abiziwe, okuqinisekisa ukuthi ukugeleza kwethu kusebenza njengoba bekulindelekile.
Phezulu kwephaneli yesandla sokudla, kukhona 'Engeza kudathasethi' inkinobho. Kunoma isiphi isinyathelo sesihlahla, le nkinobho, lapho ichofozwa, izovula iphaneli efana naleli eliboniswe ngezansi, lapho ungakwazi ukwengeza okokufaka nokukhiphayo kwaleso sinyathelo ngokuqondile kudathasethi yokuhlola edalwe esigabeni sangaphambilini. Lesi isici esiwusizo kochwepheshe babantu ukuze bengeze imibuzo yabasebenzisi evame ukwenzeka kanye nezimpendulo ezinhle kudathasethi phakathi nokusebenza komenzeli okuvamile, ngaleyo ndlela kwakhiwe inqolobane yokuhlola Ukuhlehla ngomzamo omncane. Ngokuzayo, uma kukhona ukuthuthukiswa okukhulu noma ukukhishwa kuhlelo lokusebenza, isethi yedatha yokuhlehla ingaqhutshwa futhi okuphumayo okukhiqizwayo kungatholwa ngokuqhathaniswa nokuphumayo Okulindelekile (iqiniso eliyisisekelo) elirekhodwe lapha kusetshenziswa i- 'Ukunemba kwe-GT' umhloli esisidale ngesikhathi sokusethwa kwe-LLM-as-a-jaji. Lokhu kusiza ukuthola ukukhukhuleka kwe-LLM (noma i-ejenti ukukhukhuleka) kusenesikhathi futhi uthathe izinyathelo zokulungisa.
Nansi enye yezindlela zokuhlola (Ukufingqa) kulo mkhondo wohlelo lokusebenza. Umhloli uhlinzeka ngesizathu samaphuzu angu-0.4 anqume ukuthi le mpendulo iyiyo.
Izikolo
Inketho ye-Scores ku-Langfuse ibonisa uhlu lwazo zonke izivivinyo ezivela kubahloli abahlukahlukene abasebenzayo kanye nezikolo zabo. Okubaluleke kakhulu ideshibhodi Yezibalo, lapho kungakhethwa khona amaphuzu amabili futhi amamethrikhi afana nokuchezuka okumaphakathi nokujwayelekile kanye nemigqa yethrendi ingabukwa.
Ukuhlolwa kokuhlehla
Ngama-Datasets, silungele ukuqalisa ukuhlolwa kokuhlehla sisebenzisa inqolobane yekesi yokuhlola yemibuzo kanye nemiphumela elindelekile. Sigcine imibuzo emi-4 kudathasethi yethu yokuhlehla, ngengxube yemibuzo yobuchwepheshe, yokukhokha kanye 'Yokubili'.
Kulokhu, singasebenzisa ikhodi enamathiselwe ethola idathasethi efanelekile bese iqhuba isilingo. Wonke ama-run okuhlola afakiwe kanye nesilinganiso samaphuzu. Singabuka umphumela wokuhlolwa okukhethiwe nge Ukufingqa, Ukunemba kwe-GT kanye Nokufaneleka amaphuzu ekesi ngalinye lokuhlola kudeshibhodi eyodwa. Futhi njengoba kudingeka, umkhondo onemininingwane ungafinyelelwa ukuze ubone isizathu somphumela.
Ungabuka ikhodi lapha.
from langfuse import get_client
from langfuse.openai import OpenAI
from langchain_openai import AzureChatOpenAI
from langfuse import Langfuse
import os
# Initialize client
from dotenv import load_dotenv
load_dotenv(override=True)
langfuse = Langfuse(
host="
public_key=os.environ["LANGFUSE_PUBLIC_KEY"] ,
secret_key=os.environ["LANGFUSE_SECRET_KEY"]
)
llm = AzureChatOpenAI(
azure_deployment=os.environ.get("AZURE_OPENAI_DEPLOYMENT_NAME"),
api_version=os.environ.get("AZURE_OPENAI_API_VERSION", "2025-01-01-preview"),
temperature=0.2,
)
# Define your task function
def my_task(*, item, **kwargs):
question = item.input['ticket']
response = llm.invoke([{"role": "user", "content": question}])
raw = response.content.strip().lower()
return raw
# Get dataset from Langfuse
dataset = langfuse.get_dataset("Regression")
# Run experiment directly on the dataset
result = dataset.run_experiment(
name="Production Model Test",
description="Monthly evaluation of our production model",
task=my_task # see above for the task definition
)
# Use format method to display results
print(result.format())
Okuthathwayo Okubalulekile
- Ukubonakala kwe-AI akudingeki kube yikhodi-nzima.
Amakhono amaningi okuhlola, ukulandelela, kanye nokuhlehla kokuhlolwa kwama-ejenti e-LLM anganikwa amandla ngokucupha esikhundleni sekhodi yangokwezifiso, kwehlise kakhulu ukuthuthukiswa nomzamo wokulungisa. - Ukugeleza komsebenzi wokuhlola okucebile kungachazwa ngokudalula.
Amakhono afana namagoli we-LLM-as-a-Judge (ukuhambisana, ukufingqa, ukubona izinto ezingekho, ukunemba kweqiniso eliphansi), imephu eguquguqukayo, kanye nemiyalo yokuhlola kumiswa ngokuqondile endaweni yokubonakala—ngaphandle kokubhala umqondo wokuhlola owenziwe kahle. - Amasethi edatha nokuhlolwa kokuhlehla kuyizici zokulungisa kuqala.
Amakhosombe wecala lokuhlola, ukugijima kwedathasethi, nokuqhathanisa kweqiniso eliyisisekelo kungase kusethwe futhi kusetshenziswe kabusha nge-UI noma ukumisa okulula, okuvumela amaqembu ukuthi aqhube ukuhlolwa kokuhlehla kuzo zonke izinguqulo zomenzeli anekhodi encane eyengeziwe. - Ukubonakala okuphelele kwe-MELT kuphuma “ngaphandle kwebhokisi.”
Amamethrikhi (ukubambezeleka, ukusetshenziswa kwamathokheni, izindleko), imicimbi (i-LLM namakholi wamathuluzi), amalogi, nokulandelelwa kuthathwe ngokuzenzakalelayo futhi kuhlotshaniswe, kugwenywa isidingo sokusebenza mathupha kuwo wonke ama-ejenti wokugeleza komsebenzi. - Izinsimbi zomculo ezincane, ukubonakala okuphezulu.
Ngokuhlanganiswa kwe-SDK engasindi, amaqembu athola ukubonakala okujulile ezindleleni zokusebenzisa ama-ejenti amaningi, imiphumela yokuhlola, namathrendi okusebenza—okukhulula onjiniyela ukuze bagxile kungqondongqondo ye-ejenti kunokubonakala kwamapayipi.
Isiphetho
Njengoba ama-agent e-LLM eba yinkimbinkimbi kakhulu, ukubonakala akusakhetheki. Ngaphandle kwakho, amasistimu ama-ejenti amaningi asheshe aguquke abe amabhokisi amnyama okunzima ukuwahlola, ukuwasusa, nokuthuthuka.
Inkundla yokubonakala ye-AI isusa lo mthwalo kubathuthukisi kanye nekhodi yohlelo lokusebenza. Ukusebenzisa a Ikhodi encane, indlela yokumisa-yokuqalaamaqembu angakwazi ukunika amandla ukuhlolwa kwe-LLM-as-a-Jaji, ukuhlolwa kokuhlehla, nokubonakala okuphelele kwe-MELT ngaphandle kokwakha nokugcina amapayipi angokwezifiso. Lokhu akunciphisi umzamo wobunjiniyela kuphela kodwa futhi kusheshisa indlela esuka ku-prototype iye ekukhiqizeni.
Ngokwamukela umthombo ovulekile, inkundla yohlaka-agnostic njengeLangfuse, amaqembu azuza a umthombo owodwa weqiniso ekusebenzeni kwe-ejenti—okwenza amasistimu e-AI abe lula ukwethenjwa, aguquke, futhi asebenze esikalini.
Ufuna ukwazi okwengeziwe? Uhlelo lokusebenza lwe-ejenti Yesevisi Yekhasimende olwethulwa lapha lulandela iphethini yesakhiwo somphathi nesisebenzi, okuyinto akunjalo sebenza ku-CrewAI. Funda ukuthi kanjani ukubonwa kwangisiza ukulungisa le nkinga eyaziwa kahle ngenqubo yokuhlelwa kwesigaba somphathi nesisebenzi se-CrewAI, ngokulandelela izimpendulo ze-ejenti esinyathelweni ngasinye futhi ngizicwengisise ukuze i-orchestration isebenze ngendlela efanele. Ukuhlaziya okuphelele lapha: Kungani I-CrewAI's Manager-Worker Architecture Yehluleka – nokuthi Ilungiswa Kanjani
Xhumana nami futhi wabelane ngamazwana akho ku-www.linkedin.com/in/partha-sarkar-lets-talk-AI
Zonke izithombe nedatha esetshenziswe kulesi sihloko kukhiqizwa ngokwenziwa. Izibalo nekhodi edalwe yimi
