starcoach-hackathon/zarinpal-analytics-backend
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix(growth): harden AI jobs and cost controls

Browse Source
This commit is contained in:
Amirhossein Khalili
2026-08-21 01:48:47 +03:30
parent 86e9d7c5f9
commit a15c04cac1
13 changed files with 366 additions and 78 deletions
Download Patch File Download Diff File Expand all files Collapse all files

103
internal/growthreport/avalai.go
View File

@@ -7,6 +7,7 @@ import (
"errors"
"fmt"
"io"
"math/big"
"net/http"
"strings"
"time"
@@ -60,17 +61,19 @@ func (c *AvalAIClient) GenerateWithTools(ctx context.Context, evidence json.RawM
"text": map[string]any{"format": map[string]any{"type": "json_schema", "name": "growth_copy", "strict": true, "schema": aiCopySchema()}},
}
var last error
var total Usage
for attempt := 0; attempt < 2; attempt++ {
copy, usage, transient, err := c.workflow(ctx, base, evidence, known, executor)
total = mergeUsage(total, usage)
if err == nil {
return copy, usage, nil
return copy, total, nil
}
last = err
if !transient || ctx.Err() != nil {
break
}
}
return AICopy{}, Usage{}, last
return AICopy{}, total, last
}
type responseEnvelope struct {
@@ -99,9 +102,9 @@ type outputItem struct {
func (c *AvalAIClient) workflow(ctx context.Context, base map[string]any, evidence json.RawMessage, known map[string]struct{}, executor GrowthAnalyticsTools) (AICopy, Usage, bool, error) {
firstBody := cloneRequest(base)
firstBody["input"] = string(evidence)
first, latency, transient, err := c.do(ctx, firstBody)
first, usage, transient, err := c.do(ctx, firstBody)
if err != nil {
return AICopy{}, Usage{}, transient, err
return AICopy{}, usage, transient, err
}
calls := make([]outputItem, 0, 4)
for _, raw := range first.Output {
@@ -110,9 +113,6 @@ func (c *AvalAIClient) workflow(ctx context.Context, base map[string]any, eviden
calls = append(calls, item)
}
}
usage := Usage{InputTokens: first.Usage.InputTokens, OutputTokens: first.Usage.OutputTokens, LatencyMS: latency}
usage.ProviderRequestID = first.ProviderRequestID
usage.EstimatedCost = costString(first.EstimatedCost)
final := first
if len(calls) > 0 {
if executor == nil {
@@ -141,19 +141,11 @@ func (c *AvalAIClient) workflow(ctx context.Context, base map[string]any, eviden
}
secondBody := cloneRequest(base)
secondBody["input"] = input
second, secondLatency, secondTransient, err := c.do(ctx, secondBody)
second, secondUsage, secondTransient, err := c.do(ctx, secondBody)
usage = mergeUsage(usage, secondUsage)
if err != nil {
return AICopy{}, usage, secondTransient, err
}
usage.InputTokens += second.Usage.InputTokens
usage.OutputTokens += second.Usage.OutputTokens
usage.LatencyMS += secondLatency
if second.ProviderRequestID != "" {
usage.ProviderRequestID = second.ProviderRequestID
}
if cost := costString(second.EstimatedCost); cost != "" {
usage.EstimatedCost = cost
}
final = second
for _, raw := range second.Output {
var item outputItem
@@ -169,44 +161,76 @@ func (c *AvalAIClient) workflow(ctx context.Context, base map[string]any, eviden
return copy, usage, false, nil
}
func (c *AvalAIClient) do(parent context.Context, bodyValue map[string]any) (responseEnvelope, int64, bool, error) {
func (c *AvalAIClient) do(parent context.Context, bodyValue map[string]any) (responseEnvelope, Usage, bool, error) {
body, err := json.Marshal(bodyValue)
if err != nil {
return responseEnvelope{}, 0, false, fmt.Errorf("encoding AvalAI request: %w", err)
return responseEnvelope{}, Usage{}, false, fmt.Errorf("encoding AvalAI request: %w", err)
}
ctx, cancel := context.WithTimeout(parent, c.config.Timeout)
defer cancel()
req, err := http.NewRequestWithContext(ctx, http.MethodPost, strings.TrimRight(c.config.BaseURL, "/")+"/v1/responses", bytes.NewReader(body))
if err != nil {
return responseEnvelope{}, 0, false, fmt.Errorf("creating AvalAI request: %w", err)
return responseEnvelope{}, Usage{}, false, fmt.Errorf("creating AvalAI request: %w", err)
}
req.Header.Set("Authorization", "Bearer "+c.config.APIKey)
req.Header.Set("Content-Type", "application/json")
started := time.Now()
resp, err := c.client.Do(req)
if err != nil {
return responseEnvelope{}, 0, true, fmt.Errorf("calling AvalAI: %w", err)
return responseEnvelope{}, Usage{ProviderRequests: 1, LatencyMS: time.Since(started).Milliseconds()}, true, fmt.Errorf("calling AvalAI: %w", err)
}
defer resp.Body.Close()
responseBody, err := io.ReadAll(io.LimitReader(resp.Body, maxAvalAIResponseBytes+1))
if err != nil {
return responseEnvelope{}, 0, true, fmt.Errorf("reading AvalAI response: %w", err)
return responseEnvelope{}, Usage{ProviderRequests: 1, LatencyMS: time.Since(started).Milliseconds(), ProviderRequestID: resp.Header.Get("X-Request-ID")}, true, fmt.Errorf("reading AvalAI response: %w", err)
}
if len(responseBody) > maxAvalAIResponseBytes {
return responseEnvelope{}, 0, false, errors.New("avalai: response too large")
}
if resp.StatusCode < 200 || resp.StatusCode >= 300 {
return responseEnvelope{}, 0, resp.StatusCode == 429 || resp.StatusCode >= 500, fmt.Errorf("avalai: HTTP %d", resp.StatusCode)
return responseEnvelope{}, Usage{ProviderRequests: 1, LatencyMS: time.Since(started).Milliseconds(), ProviderRequestID: resp.Header.Get("X-Request-ID")}, false, errors.New("avalai: response too large")
}
var result responseEnvelope
if err := json.Unmarshal(responseBody, &result); err != nil {
return responseEnvelope{}, 0, false, fmt.Errorf("decoding AvalAI response: %w", err)
decodeErr := json.Unmarshal(responseBody, &result)
result.ProviderRequestID = resp.Header.Get("X-Request-ID")
usage := Usage{InputTokens: result.Usage.InputTokens, OutputTokens: result.Usage.OutputTokens, LatencyMS: time.Since(started).Milliseconds(), ProviderRequests: 1, ProviderRequestID: result.ProviderRequestID, EstimatedCost: costString(result.EstimatedCost)}
if resp.StatusCode < 200 || resp.StatusCode >= 300 {
return result, usage, resp.StatusCode == 429 || resp.StatusCode >= 500, fmt.Errorf("avalai: HTTP %d", resp.StatusCode)
}
if decodeErr != nil {
return responseEnvelope{}, usage, false, fmt.Errorf("decoding AvalAI response: %w", decodeErr)
}
if result.Status != "completed" {
return responseEnvelope{}, 0, false, fmt.Errorf("avalai: incomplete response status %q", result.Status)
return result, usage, false, fmt.Errorf("avalai: incomplete response status %q", result.Status)
}
result.ProviderRequestID = resp.Header.Get("X-Request-ID")
return result, time.Since(started).Milliseconds(), false, nil
return result, usage, false, nil
}
func mergeUsage(total, next Usage) Usage {
total.InputTokens += next.InputTokens
total.OutputTokens += next.OutputTokens
total.LatencyMS += next.LatencyMS
total.ProviderRequests += next.ProviderRequests
total.ReservedTokens += next.ReservedTokens
if next.ProviderRequestID != "" {
total.ProviderRequestID = next.ProviderRequestID
}
total.EstimatedCost = addDecimalStrings(total.EstimatedCost, next.EstimatedCost)
return total
}
func addDecimalStrings(left, right string) string {
if left == "" {
return right
}
if right == "" {
return left
}
a, okA := new(big.Rat).SetString(left)
b, okB := new(big.Rat).SetString(right)
if !okA || !okB {
return left
}
text := new(big.Rat).Add(a, b).FloatString(12)
text = strings.TrimRight(strings.TrimRight(text, "0"), ".")
return text
}
func costString(raw json.RawMessage) string {
@@ -287,6 +311,9 @@ func ValidateAICopy(copy AICopy, known map[string]struct{}) error {
return errors.New("avalai: numeric claim rejected")
}
}
if containsQuantityWord(field) {
return errors.New("avalai: written numeric claim rejected")
}
lower := strings.ToLower(field)
for _, forbidden := range []string{"payer", "card", "رمز", "شماره کارت", "sql", "merchant_id"} {
if strings.Contains(lower, forbidden) {
@@ -306,6 +333,20 @@ func ValidateAICopy(copy AICopy, known map[string]struct{}) error {
return nil
}
func containsQuantityWord(value string) bool {
quantityWords := map[string]struct{}{
"zero": {}, "one": {}, "two": {}, "three": {}, "four": {}, "five": {}, "six": {}, "seven": {}, "eight": {}, "nine": {}, "ten": {}, "eleven": {}, "twelve": {}, "thirteen": {}, "fourteen": {}, "fifteen": {}, "sixteen": {}, "seventeen": {}, "eighteen": {}, "nineteen": {}, "twenty": {}, "thirty": {}, "forty": {}, "fifty": {}, "sixty": {}, "seventy": {}, "eighty": {}, "ninety": {}, "hundred": {}, "thousand": {}, "million": {}, "billion": {}, "percent": {}, "percentage": {}, "double": {}, "triple": {}, "half": {},
"صفر": {}, "یک": {}, "دو": {}, "سه": {}, "چهار": {}, "پنج": {}, "شش": {}, "هفت": {}, "هشت": {}, "نه": {}, "ده": {}, "یازده": {}, "دوازده": {}, "سیزده": {}, "چهارده": {}, "پانزده": {}, "شانزده": {}, "هفده": {}, "هجده": {}, "نوزده": {}, "بیست": {}, "سی": {}, "چهل": {}, "پنجاه": {}, "شصت": {}, "هفتاد": {}, "هشتاد": {}, "نود": {}, "صد": {}, "هزار": {}, "میلیون": {}, "میلیارد": {}, "درصد": {}, "برابر": {}, "نصف": {},
}
for _, word := range strings.Fields(strings.ToLower(value)) {
word = strings.Trim(word, ".,;:!?،؛؟٪%()[]{}«»\"'")
if _, found := quantityWords[word]; found {
return true
}
}
return false
}
func growthTools() []map[string]any {
names := []string{"get_growth_context", "get_event_evidence", "get_peer_context", "get_campaign_playbook"}
tools := make([]map[string]any, 0, len(names))