Yeqa uye kokuyinhloko
Izifundo ze-AI

Ukwakha Kusukela Ekuqaleni Isevisi Yamakhasimende Ye-E-commerce Esebenzisa i-Adaptive-RAG: Ukusesha Okuzivumelanisa Nesimo, Ukuhlonza Inhloso Namathuluzi Ama-oda

Lesi sifundo sizokuholela ekusebenziseni isevisi yamakhasimende ye-e-commerce esebenzisa i-Adaptive-RAG endaweni yangakini, uqonde ukusesha okuguqukayo, ukugcwalisa ubufakazi ngemijikelezo eminingi, ukuqinisekisa ukwethembeka, ukuhlonza izinhloso nge-Jev ne-DeepSeek, futhi uzame ukuphendula imibuzo ngezinqubomgomo, ukuhlola ama-oda, ukuqinisekiswa komuntu, ukuhlola ukusebenza nokusatshalaliswa nge-Docker.

Cishe imizuzu engu-42 yokufunda
Isevisi yamakhasimende ye-e-commerce ye-Adaptive-RAG: ukusesha, ukuhlonza izinhloso namathuluzi ama-oda

Isingeniso sephrojekthi

I-Adaptive-RAG E-commerce Customer Service iwuhlelo lwesevisi yamakhasimende oluhlakaniphile olugxile ekuphenduleni imibuzo yezinqubomgomo enemibandela eminingi. Ihlose ukuxazulula izinkinga ezifana nobufakazi obunganele, izimpendulo ezingenasisekelo esicacile, kanye nemibuzo eyinkimbinkimbi engakwazi ukuqedwa ngokusesha okukodwa. I-backend iqhutshwa yi-FastAPI ne-LangGraph, kanti i-frontend isebenzisa i-React. Uhlelo lukhetha ngamandla phakathi kokuphendula ngokuqondile, ukubuyisa ulwazi, ukusesha okwengeziwe, ukusebenzisa amathuluzi, ukuqinisekiswa ngumuntu, nokwenqaba ukuphendula, kuye ngenhloso yomsebenzisi nobunkimbinkimbi bombuzo.

Le phrojekthi ilungele ukuboniswa endaweni yangakini, ukufunda ukuhamba komsebenzi kwe-Adaptive RAG, kanye nokuqinisekisa ukusebenza kobunjiniyela. Isekela izingcaphuno zemithombo yezinqubomgomo, imibuzo yama-oda okubukisa, imisebenzi elindele ukuqinisekiswa, inkumbulo yeseshini, nezimpendulo ze-SSE, kodwa idatha yama-oda neyokulethwa akuyona idatha yebhizinisi yangempela.

I-API yamanje yamukela ngokuqondile i-user_id ethunyelwa iklayenti, kodwa ayikaqinisekisi ubunikazi bangempela. Ungayifaki le phrojekthi ngokuqondile ku-inthanethi yomphakathi ngaphandle kokuqale uhlanganise ukuqinisekiswa kobunikazi okuthembekile, i-TLS nokulawulwa kokufinyelela.

Indlela uhlelo oluphatha ngayo isicelo

Ngemva kokuba umsebenzisi ethumele umbuzo ngekhasi le-Web noma nge-HTTP API, isicelo sidlula ngokulandelana kusihlungi sokufaka, ekuhlonzweni kwenhloso, nasekuqondisweni kobunkimbinkimbi. Uhlelo bese lukhetha igatsha elifanele lokucubungula, lugcine ngokukhiqiza impendulo, luhlole ukwethembeka kwayo, lubuyekeze inkumbulo yeseshini, bese lubuyisa izingcaphuno.

Umsebenzisi → Web / HTTP API → Isihlungi sokufaka
     → Ukuhlonza inhloso → Ukuqondisa ubunkimbinkimbi
       ├─ Ukuphendula ngokuqondile
       ├─ Ukusesha → Ukuhlelwa kabusha → Ukusesha okwengeziwe lapho kudingeka
       ├─ Ukuhlela → Ukusebenzisa ithuluzi / Ukuqinisekiswa ngumuntu
       └─ Ukwenqaba ukuphendula
     → Ukukhiqiza nokuhlola ukwethembeka → Inkumbulo yeseshini → Impendulo nezingcaphuno

Ukuhlonza inhloso kungasebenzisa i-DeepSeek ezenzakalelayo, noma kushintshelwe ku-TypeSafe Jev. Qaphela ukuthi i-Jev ishintsha kuphela inodi yokuhlukanisa inhloso; ukunquma ubunkimbinkimbi, ukubhala kabusha umbuzo, ukuhlela umsebenzi, ukukhiqiza impendulo, ukuhlola ukwethembeka, nokufingqa inkumbulo kusaqhutshwa yi-DeepSeek.

Amakhono abalulekile

  • Ukuqondisa okuzivumelanayo: Kuhlanganisa inhloso yebhizinisi nobunkimbinkimbi bombuzo ukuze kukhethwe ukuphendula ngokuqondile, ukusesha okukodwa, ukusesha okunemijikelezo eminingi, ukusebenzisa ithuluzi, noma ukwenqaba ukuphendula.
  • Ukusesha okuxubile: Kuhlanganisa ukubuyisa nge-BM25 nokubuyisa ngamavekhtha aminyene, kusebenzise i-RRF ukuhlanganisa imiphumela, futhi kungavula ukuhlelwa kabusha nge-Cross-Encoder.
  • Ukwandisa ubufakazi: Kusekela ukubhalwa kabusha kombuzo, ukusesha okwengeziwe lapho ubufakazi bunganele, ukuhlolwa kokwethembeka kwempendulo, nokuzilungisa.
  • Ama-backend amabili enhloso: Isebenzisa i-DeepSeek ngokuzenzakalelayo, kodwa ingasebenzisa isixhumi esibonakalayo esihlelekile se-choice se-Jev ukuhlonza izinhlobo ezingu-14 ezichazwe kusengaphambili zezinhloso zebhizinisi.
  • Ukucubungula amadokhumenti: Isekela i-Markdown ngokuqondile; ngemva kokufaka okuncikile okukhethwayo namathuluzi esistimu, ingahlaziya amadokhumenti e-PDF, e-Office nawe-OCR.
  • Ukuhlukanisa okuhlelekile: Umbhalo ojwayelekile usebenzisa iwindi lezinhlamvu ezingu-500 nokugqagqana kwezinhlamvu ezingu-60 ngokuzenzakalelayo; amathebula abonwayo ahlukaniswa abe yizingxenye zawo, kugcinwe imethadatha efana nesihloko, isahluko nomthombo.
  • Amathuluzi ebhizinisi: Ihlinzeka ngamathuluzi okubonisa ama-oda, okuhlanganisa ukuhlolwa kwezimvume, ukwehlisa ukusebenza, ukuqinisekiswa ngumuntu, nezindlela ze-idempotency, futhi ingaxhuma namathuluzi e-MCP.
  • Amakhono obunjiniyela: Kuhlanganisa ukucindezela inkumbulo yeseshini, izingodo ezihlelekile, amamethrikhi e-Prometheus, ukuhlolwa kwempilo, nokuhlolwa okuzenzakalelayo.

Ukulungiselela indawo yokusebenza

Izidingo eziyisisekelo

  • Python 3.11 noma inguqulo entsha.
  • Uma udinga ukusebenzisa i-Web frontend, sebenzisa i-Node.js 20.19+ noma 22.12+, futhi ulungiselele i-npm.
  • Imodi yangakini ezenzakalelayo isebenzisa i-SQLite nezingxenye zokusesha ezisememori; akudingeki uqale i-Docker kusengaphambili.

Izinyathelo zokufaka ku-Windows PowerShell

Ngemva kokuklona i-repository nokungena kumkhombandlela oyimpande wephrojekthi, dala indawo ebonakalayo, faka okuncikile kokuthuthukisa, kopisha ifayela lokumisa, bese ukhiqiza idatha yokubonisa:

python -m venv .venv
.\.venv\Scripts\python.exe -m pip install -e ".[dev]"
Copy-Item .env.example .env
.\.venv\Scripts\python.exe -m scripts.seed_demo

Uma i-.env isivele ikhona kumkhombandlela oyimpande wephrojekthi, gcina ifayela elikhona bese weqa umyalo wokukopisha. Bese uqala i-FastAPI:

.\.venv\Scripts\python.exe -m uvicorn app.asgi:app --reload --host 127.0.0.1 --port 8000

macOS ne-Linux

Kuma-macOS noma e-Linux, ungadala futhi uvule indawo ebonakalayo, bese usebenzisa imiyalo yokufaka ehambisanayo:

python3 -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"
cp .env.example .env
python -m scripts.seed_demo
python -m uvicorn app.asgi:app --reload --host 127.0.0.1 --port 8000

Ukumisa amamodeli namasevisi endawo

Vula i-.env ekumkhombandlela oyimpande wephrojekthi. Lezi yizilungiselelo okufanele uzihlole kuqala:

  • DEEPSEEK_API_KEY: Ivula ukuhamba komsebenzi kwe-DeepSeek nokuhlukanisa inhloso okuzenzakalelayo. Izingcingo zangempela ze-API zingadala izindleko.
  • INTENT_BACKEND: Ingaba yi-deepseek ezenzakalelayo noma i-jev ekhethwayo.
  • JEV_API_KEY: Iyadingeka uma kuvulwa ukuhlukanisa inhloso kwe-Jev.
  • JEV_MODEL: Inani elizenzakalelayo lamanje lephrojekthi ngu-jev-1.13.0.
  • DATABASE_URL: Ngokuzenzakalelayo isebenzisa i-sqlite+aiosqlite:///./data/local/app.db.
  • KNOWLEDGE_DIR: Inani elizenzakalelayo ngu-knowledge; amadokhumenti olwazi akuyo afakwa kusizindalwazi ekuqaleni.
  • LOCAL_EMBEDDING_ENABLED: Ilawula ukusebenza kwamavekhtha encazelo asebenza endaweni yangakini.
  • CROSS_ENCODER_ENABLED: Ilawula ukuhlelwa kabusha kwendawo yangakini.
  • REDIS_URL ne-QDRANT_URL: Kumisa izinsiza zangaphandle ozikhethela zona zokugcina isikhashana namavekhtha.

Uma ungamisi ukhiye we-DeepSeek, uhlelo lusengasebenzisa umqondo wokubonisa ongaxhunyiwe ku-inthanethi, kodwa ikhwalithi yezimpendulo izohluka kumodi yemodeli yangempela. Uhlu oluphelele lweziguquguquko lutholakala ku-Inkomba yokumisa naku-ifayela lesibonelo sendawo.

Ukuqalisa isixhumi esibonakalayo sengxoxo se-React

Gcina inqubo ye-backend isebenza, bese kwenye i-terminal uye kumkhombandlela we-frontend:

cd frontend
npm ci
npm run dev

Vakashela http://127.0.0.1:5173 ukuze uvule ikhasi lengxoxo. Iseva yokuthuthukisa izodlulisela izicelo ze-/api echwebeni lendawo elingu-8000.

I-frontend isekela izingcaphuno zemithombo, uhlu lweseshini lwendawo, namakhadi okuqinisekisa izenzo. I-backend ithumela okuqukethwe ngezingxenye nge-SSE ngemva kokuba impendulo iqediwe ukuhlolwa. Uma usebenzisa i-npm run build ukukhiqiza i-frontend/dist, i-FastAPI izohlinzeka ngekhasi le-frontend lisuka endleleni eyimpande lapho ibona lowo mkhombandlela.

Ukusetshenziswa okuyisisekelo: Ukubuza ngenqubomgomo yokubuyisa nokushintsha

Qala ngokuqinisekisa isimo sesevisi:

Okulandelayo kusebenzisa i-PowerShell ukuthumela umbuzo onezimo eziningi zenqubomgomo kusixhumi esibonakalayo sengxoxo:

$payload = @{
    message = "定制马克杯过了七天，但是有质量问题，可以退吗？"
    user_id = "u_demo"
} | ConvertTo-Json

Invoke-RestMethod `
    -Uri "http://127.0.0.1:8000/api/v1/chat" `
    -Method Post `
    -ContentType "application/json; charset=utf-8" `
    -Body ([System.Text.Encoding]::UTF8.GetBytes($payload))

Uhlelo luzoqala luhlonze inhloso ehlobene nokubuyisa nokushintsha, bese lunquma ngokobunkimbinkimbi bombuzo ukuthi ludinge yini ukusesha amadokhumenti enqubomgomo. Uma ubufakazi bokuqala bunganele, ukuhamba komsebenzi kungabhala kabusha umbuzo bese kwenza ukusesha okwengeziwe; ngemva kokukhiqiza impendulo, luzophinde luhlole ukuthi impendulo ithembekile yini ebufakazini obuseshiwe, bese lubuyisa izingcaphuno zemithombo ezitholakalayo.

Ngaphandle kwe-HTTP API, ungasebenzisa lo myalo olandelayo ngesikhathi i-backend isebenza, usebenzise iklayenti le-terminal:

python -m app.cli

Ukusetshenziswa okuyisisekelo: Ukubuza nokusebenzisa ama-oda okubukisa

I-scripts.seed_demo idala ngokusobala ama-oda okubukisa. Ukuqalisa i-API ngokwako akubhali le datha ngokuzenzakalelayo, ngakho ungakweqi ukuqaliswa kwedatha lapho usebenzisa okokuqala. Umsebenzisi wokubonisa ngu-u_demo, futhi ama-oda atholakalayo yilawa:

  • ORD10001: Kuthunyelwe, umkhiqizo ungama-headphone anciphisa umsindo.
  • ORD10002: Kwamukelwe, umkhiqizo uyinkomishi eyenziwe ngokwezifiso.
  • ORD10003: Kukhokhiwe kodwa akukathunyelwa, umkhiqizo yi-T-shirt.

Ngokwesibonelo, faka 取消订单 ORD10003 ekhasini lengxoxo. Lapho uhlelo lusebenzisa isenzo esidinga ukuqinisekiswa, lungabuyisa i-pending_action. Isicelo sokuqinisekisa kufanele sibuyise ne-session_id esiphendulweni, futhi sifake i-pending_action.action_id ku-confirm_action_id. Ikhasi le-Web lisebenzisa ikhadi lokuqinisekisa ukusiza ukuqedela le nqubo.

Izenzo ezilindele ukuqinisekiswa zigcinwa ngokuzenzakalelayo ngaphakathi kwenqubo yeseva imizuzu engu-15. Ngemva kokuqalisa kabusha iseva noma ukushintshela kwenye inqubo, kungase kudingeke uqale isenzo futhi. Ukuze ubone izinkambu nefomethi yesixhumi esibonakalayo, bheka Inkomba ye-API.

Ukumisa okuthuthukile: Ukuvula ukuhlonza inhloso kwe-Jev

Ukuze ushintshele inodi yokuhlukanisa inhloso ku-TypeSafe Jev, engeza noma uguqule lezi zilungiselelo ku-.env yendawo:

INTENT_BACKEND=jev
JEV_API_KEY=your_jev_api_key
JEV_ENDPOINT=https://api.typesafe.ai/v1/systemone
JEV_MODEL=jev-1.13.0
JEV_TIMEOUT_SECONDS=10
JEV_MAX_RETRIES=2
JEV_INTENT_APPLY_OVERLAY=true

DEEPSEEK_API_KEY=your_deepseek_api_key

Ngemva kokulondoloza, qala kabusha i-backend. I-Jev ixhunywa ngesixhumi esibonakalayo se-HTTP futhi ayidingi i-SDK ekhethekile. Imiphumela ebuyiswayo iqinisekiswa nge-Pydantic, bese iguqulelwa kumphumela wokuhlukanisa ohlanganisiwe.

Uma ukhiye we-Jev ungekho, isicelo sihluleka, noma impendulo ingavumelekile, uhlelo lubuyela ku-IntentClassifier ekhona. Uchungechunge lokubuyela emuva lusebenzisa kuqala i-DeepSeek emisiwe; uma lungaxhunyiwe ku-inthanethi noma ukubizwa kwemodeli kwehluleka, luyaqhubeka nokusebenzisa ukuhlukanisa okusekelwe emithethweni. Amaphutha angaphindwa asebenzisa ukuhlehla kwe-exponential, futhi i-JEV_MAX_RETRIES ekusetshenzisweni kwamanje imele inani eliphezulu lemizamo.

JEV_INTENT_APPLY_OVERLAY ilawula ukulungiswa kwemithetho ngemva kokuhlukanisa ngempumelelo kwe-Jev, futhi ivuliwe ngokuzenzakalelayo. Le nketho ayilawuli ungqimba lokulungisa lwe-classifier yokubuyela emuva ngokwayo. Ukuze ubuyisele ukuhlukanisa inhloso okuzenzakalelayo, shintsha nje i-INTENT_BACKEND uyibuyisele ku-deepseek.

Ukucushwa okuthuthukisiwe: ukuthuthukisa ukusesha nokuhlaziywa kwamadokhumenti

Faka okuncike kuma-vector asekhaya kanye nokuhlela kabusha imiphumela

.\.venv\Scripts\python.exe -m pip install -e ".[rerank]"

Okuncike ku-rerank kusetshenziselwa amamodeli e-dense vector asekhaya kanye ne-Cross-Encoder. Imodeli ye-vector ezenzakalelayo yi-BAAI/bge-small-zh-v1.5, kanti imodeli ezenzakalelayo yokuhlela kabusha yi-BAAI/bge-reranker-base. Lapho ilayishwa okokuqala, izisindo zemodeli zingalandwa; uma imodeli ingatholakali, uhlelo luzokwehla lusebenzise enye indlela, ngakho imiphumela ngeke ilingane nokucushwa okuphelele kwemodeli.

Faka okuncike kukho ekwandiseni ukuhlaziywa kwamadokhumenti

.\.venv\Scripts\python.exe -m pip install -e ".[docs]"

Okuncike ku-docs kusetshenziselwa ukuhlaziywa kwamadokhumenti nge-Unstructured. Lapho kusetshenzwa ama-PDF, i-OCR namafayela e-Office, kungase kudingeke futhi ukuthi kufakwe i-Poppler, i-Tesseract namaphakheji olimi lwesiShayina, kanye ne-LibreOffice. Izidingo eziningiliziwe zitholakala ku-Incazelo yokuhlaziya nokusesha.

I-ID yesiqephu solwazi ikhiqizwa ngokusekelwe endleleni yedokhumenti kanye nenombolo yesiqephu. Uma indlela nemiphumela yokuhlukanisa kungashintshi, ukufaka idatha kaningi kungagcina ama-ID efana; ngemva kokuthuthwa kwefayela, ukuhlelwa kokuqukethwe noma ukushintshwa kwesu lokuhlukanisa, izinkomba zomlando akuqinisekisiwe ukuthi zizoqhubeka zikhomba okuqukethwe okufanayo.

Qalisa isevisi ephelele nge-Docker

Uma ufuna ukusebenzisa ngesikhathi esisodwa i-API, i-PostgreSQL, i-Redis, i-Qdrant kanye ne-Web frontend eyakhiwe, qala ulungiselele i-.env yendawo, bese usebenzisa:

docker compose up --build

Ngemva kokuqalisa, vakashela ku-http://127.0.0.1:8000. Iphasiwedi yesizindalwazi kanye nokuxhunywa kwamachweba ku-Compose kuhloselwe ukuboniswa kwasendaweni kuphela. Ngaphambi kokufaka isevisi ku-inthanethi, kufanele ushintshe amaphasiwedi okubonisa, ukhawulele amachweba esizindalwazi nawe-cache, uvule ukuqinisekiswa kobunikazi okuthembekile ne-TLS, futhi uhlele isu elithembekile lokuphikelela kwedatha nokusebenza.

Ukuhlola, ukuhlolwa okumile nokulinganisa

Qalisa ukuhlola kwe-backend nokuhlolwa kwekhwalithi empandeni yephrojekthi:

.\.venv\Scripts\python.exe -m pytest -q
.\.venv\Scripts\python.exe -m ruff check app eval tests
.\.venv\Scripts\python.exe -m mypy
.\.venv\Scripts\python.exe -m eval.runner --task retrieval

Imiyalo yokuhlola nokwakha i-frontend kufanele isetshenziswe ngaphakathi kwefolda ye-frontend:

npm test
npm run build

Ukucushwa kokuhlola kuhlukanisa okhiye bamamodeli endawo futhi kusebenzisa isizindalwazi sokuhlola. Ngaphambi kokusebenzisa ukuhlolwa kwamamodeli angempela ngokuzimela, hlola ukucushwa kwemodeli nenethiwekhi, ngoba ukubiza amamodeli angaphandle kungase kubangele izindleko.

Qhathanisa ukuhlukaniswa kwezinhloso kwe-Jev ne-DeepSeek

Ngemva kokucushwa kokhiye bendawo bamamodeli amabili, ungasebenzisa ukuqhathanisa ngamasampula afanayo nangaphansi kwemibandela efanayo ye-overlay:

.\.venv\Scripts\python.exe -m eval.compare_intent --backend both --limit 10 --no-overlay --output data/local/intent-comparison.json

Umbiko uhlanganisa ukunemba, ukubambezeleka kwe-P50 ne-P95, inani lokubuyela kwenye indlela, kanye nenani lezikhathi zokusebenzisa imithetho ye-overlay. I---no-overlay ivala ukulungiswa kwemithetho kwabahleli bezigaba bobabili, ukuze kube lula ukubona imiphumela yokuhlukanisa yemodeli uqobo; ukususa le parameter kukuvumela ukuba uqhathanise imiphumela ngemva kokufaka i-overlay.

Idathasethi yamanje yezinhloso iyidathasethi yokuthuthukisa; ukunemba kuhlanganisa amasampula abuyele kwenye indlela, ngakho akumele kuthathwe ngokuqondile njengamandla okujwayela emodeli ezimele ye-Jev. Ukuhlolwa kwangaphambilini nakho akusho ukuthi inguqulo yamanje isiqinisekiswe kabusha, futhi akusona isiqinisekiso semiphumela yebhizinisi langempela.

Izincomo zobunjiniyela ezisebenzayo

  1. Qala ngokuqinisekisa ukugeleza kwimodi ezenzakalelayo yasendaweni: I-SQLite nezingxenye zokusesha ezisememori zanele ukwenza ukuboniswa okuyisisekelo; akudingekile ukuthi uqale ngokufaka isizindalwazi sangaphandle nesevisi yama-vector.
  2. Hlukanisa imodeli yenhloso nemodeli yokugeleza komsebenzi: Ukushintshela ku-Jev akushintshi amanye ama-node e-DeepSeek. Uma ufuna ukusebenzisa ukugeleza komsebenzi okuphelele okuku-inthanethi, kusadingeka ulungiselele i-DeepSeek.
  3. Sebenzisa iziphetho zezempilo ekuxazululeni izinkinga: I-/api/v1/health isetshenziselwa ukuhlola inqubo, kanti i-/api/v1/ready ifaneleka kakhulu ukuqinisekisa ukuthi okuncikile sekulungele.
  4. Shintsha isu lokuhlukanisa ngokuqapha: Izinguquko endleleni, kokuqukethwe kombhalo noma kumapharamitha okuhlukanisa zingathinta ama-ID eziqephu nezinkomba zomlando.
  5. Unganciki esimweni sokuqinisekisa esingaphakathi kwenqubo: Imisebenzi elinde ukuqinisekiswa ezenzakalelayo isebenza imizuzu engu-15 kuphela, futhi ingalahleka ngenxa yokuqalisa kabusha noma ukushintsha phakathi kwezinqubo eziningi.
  6. Vikela idatha ebucayi: Ungathumeli i-.env, okhiye bangempela, izizindalwazi zasendaweni, amalogi, amadokhumenti alayishwe abasebenzisi noma izinto eziyimfihlo ku-GitHub.
  7. Hlola ukudluliswa kwedatha kwangaphandle: Ngaphambi kokusebenzisa imodeli yangaphandle noma i-MCP ekude, qinisekisa ukuthi okuqukethwe kwesicelo akuqukethe ulwazi lomsebenzisi okungafanele luthunyelwe ngaphandle.
  8. Hlela ukubuyekezwa kwesizindalwazi solwazi: Uhlelo lwamanje lufaka yonke idatha ekuqaleni, futhi alukabi nendlela ephelele yokususa idatha ngokwengeza.

Ukubuka ngokushesha izinkomba zephrojekthi

  • app/api: I-HTTP API ne-SSE.
  • app/agents: Igrafu yokusebenzisa ye-LangGraph nama-node okugeleza komsebenzi.
  • app/retrieval: Ukuhlaziya, ukuhlukanisa, ukubuyisa, ukuhlanganisa nokuhlela kabusha imiphumela.
  • app/providers: Ukuvumelanisa amamodeli nezinsizakalo zama-vector.
  • app/tools: Amathuluzi ebhizinisi, ukubhaliswa kwamathuluzi nokulawulwa kwezimvume.
  • app/mcp: Ukuxhumeka kwamathuluzi e-MCP.
  • app/memory: Inkumbulo yezikhathi nokucindezelwa kwayo.
  • app/guardrails: Ukuhlolwa kokokufaka nezimpendulo.
  • app/observability: Amalogi nezinkomba.
  • frontend: Isixhumi esibonakalayo sengxoxo esakhiwe nge-React ne-TypeScript.
  • knowledge: Amadokhumenti olwazi lwezinqubomgomo.
  • tests ne-eval: Ukuhlola okuzenzakalelayo, idatha yokulinganisa nabasebenzisi bokwenza.

Isifinyezo

I-Adaptive-RAG yenkonzo yamakhasimende ye-e-commerce ikhombisa uchungechunge lobunjiniyela lwenkonzo yamakhasimende ehlakaniphile oluphelele: iqala ngokuhlonza inhloso nokuhlola ubunkimbinkimbi, bese ihlanganisa ngokwesidingo ukusesha okuxubile, ukugcwalisa ubufakazi emijikelezweni eminingi, ukubiza amathuluzi, ukuqinisekiswa komuntu kanye nokuqinisekisa ukwethembeka. Ungaqala usebenzisa i-SQLite, ama-oda okub demonstrisha kanye nengqondo engaxhunyiwe ku-inthanethi ukuze uyizame ngokushesha, bese uvula kancane kancane i-DeepSeek, i-Jev, imodeli yama-vector yasendaweni, i-Cross-Encoder, ukuhlaziywa kwamadokhumenti okunwetshiwe kanye nezinsizakalo ze-Docker.

Ngaphambi kokuyisa ekusetshenzisweni kwangempela kwebhizinisi, kusadingeka kugcwaliswe ngokukhethekile ukuqinisekiswa kobunikazi, ukuhlanganiswa nohlelo lwangempela lwama-oda, ukuphikelela kwesimo sokuqinisekisa, ukunakekelwa kwesizindalwazi solwazi ngokwengeza, ukuphathwa kokhiye kanye nokuphepha kwenethiwekhi kwezinga lokukhiqiza. Iphrojekthi isebenzisa i-MIT License; ukuze uthole incazelo eyengeziwe yesakhiwo nemikhawulo, bheka Isakhiwo nokugeleza kokusebenza, Idizayini yokusesha kanye Imikhawulo nomhlahlandlela.