diff --git a/src/components/admin/round/filtering-dashboard.tsx b/src/components/admin/round/filtering-dashboard.tsx
index 7f9cfa2..4828a13 100644
--- a/src/components/admin/round/filtering-dashboard.tsx
+++ b/src/components/admin/round/filtering-dashboard.tsx
@@ -1915,7 +1915,7 @@ function AwardTracksSection({ competitionId, roundId }: { competitionId: string;
           Special Award Tracks
         </CardTitle>
         <CardDescription>
-          Evaluate passed projects against special award criteria and manage shortlists
+          Award eligibility is evaluated automatically during AI filtering. Use Run Eligibility to re-evaluate.
         </CardDescription>
       </CardHeader>
       <CardContent className="space-y-3">
diff --git a/src/server/routers/filtering.ts b/src/server/routers/filtering.ts
index 841f1a4..43bd179 100644
--- a/src/server/routers/filtering.ts
+++ b/src/server/routers/filtering.ts
@@ -2,7 +2,8 @@ import { z } from 'zod'
 import { TRPCError } from '@trpc/server'
 import { Prisma, PrismaClient } from '@prisma/client'
 import { router, adminProcedure, protectedProcedure } from '../trpc'
-import { executeFilteringRules, type ProgressCallback } from '../services/ai-filtering'
+import { executeFilteringRules, type ProgressCallback, type AwardCriteriaInput, type AwardMatchResult } from '../services/ai-filtering'
+import { sanitizeUserInput } from '../services/ai-prompt-guard'
 import { logAudit } from '../utils/audit'
 import { isOpenAIConfigured, testOpenAIConnection } from '@/lib/openai'
 import { prisma } from '@/lib/prisma'
@@ -58,11 +59,38 @@ export async function runFilteringJob(jobId: string, roundId: string, userId: st
     // Get current round with config
     const currentRound = await prisma.round.findUniqueOrThrow({
       where: { id: roundId },
-      select: { id: true, name: true, configJson: true },
+      select: { id: true, name: true, configJson: true, competitionId: true },
     })
     const roundConfig = (currentRound.configJson as Record<string, unknown>) || {}
     const aiParseFiles = !!roundConfig.aiParseFiles
 
+    // Load special awards for integrated AI evaluation
+    let awardsForAI: AwardCriteriaInput[] = []
+    if (currentRound.competitionId) {
+      const rawAwards = await prisma.specialAward.findMany({
+        where: {
+          competitionId: currentRound.competitionId,
+          useAiEligibility: true,
+          criteriaText: { not: null },
+        },
+        select: { id: true, name: true, criteriaText: true },
+        orderBy: { sortOrder: 'asc' },
+      })
+      for (const a of rawAwards) {
+        if (a.criteriaText && a.criteriaText.trim().length > 0) {
+          const { sanitized } = sanitizeUserInput(a.criteriaText)
+          awardsForAI.push({
+            awardId: a.id,
+            awardName: a.name,
+            criteriaText: sanitized,
+          })
+        }
+      }
+      if (awardsForAI.length > 0) {
+        console.log(`[Filtering] Including ${awardsForAI.length} special award(s) in AI evaluation`)
+      }
+    }
+
     // Get projects in this round via ProjectRoundState
     const projectStates = await prisma.projectRoundState.findMany({
       where: {
@@ -189,7 +217,107 @@ export async function runFilteringJob(jobId: string, roundId: string, userId: st
           })
         )
       )
-    })
+
+      // Upsert AwardEligibility for PASSED projects with award matches
+      if (awardsForAI.length > 0) {
+        const awardUpserts: Prisma.PrismaPromise<unknown>[] = []
+        for (const r of batchResults) {
+          if (r.outcome !== 'PASSED' || !r.awardMatches || r.awardMatches.length === 0) continue
+          for (const am of r.awardMatches) {
+            awardUpserts.push(
+              prisma.awardEligibility.upsert({
+                where: {
+                  awardId_projectId: {
+                    awardId: am.awardId,
+                    projectId: r.projectId,
+                  },
+                },
+                create: {
+                  awardId: am.awardId,
+                  projectId: r.projectId,
+                  eligible: am.eligible,
+                  method: 'AUTO',
+                  qualityScore: am.qualityScore,
+                  aiReasoningJson: { reasoning: am.reasoning, confidence: am.confidence },
+                },
+                update: {
+                  eligible: am.eligible,
+                  method: 'AUTO',
+                  qualityScore: am.qualityScore,
+                  aiReasoningJson: { reasoning: am.reasoning, confidence: am.confidence },
+                  overriddenBy: null,
+                  overriddenAt: null,
+                  shortlisted: false,
+                  confirmedAt: null,
+                  confirmedBy: null,
+                },
+              })
+            )
+          }
+        }
+        if (awardUpserts.length > 0) {
+          await prisma.$transaction(awardUpserts)
+        }
+      }
+    }, awardsForAI)
+
+    // Auto-shortlist top-N per award and mark eligibility job as completed
+    if (awardsForAI.length > 0) {
+      // Collect all award matches from PASSED results
+      const awardMatchesByAward = new Map<string, Array<{ projectId: string; qualityScore: number }>>()
+      for (const r of results) {
+        if (r.outcome !== 'PASSED' || !r.awardMatches) continue
+        for (const am of r.awardMatches) {
+          if (!am.eligible) continue
+          const arr = awardMatchesByAward.get(am.awardId) || []
+          arr.push({ projectId: r.projectId, qualityScore: am.qualityScore })
+          awardMatchesByAward.set(am.awardId, arr)
+        }
+      }
+
+      // Load shortlistSize per award
+      const awardIds = awardsForAI.map((a) => a.awardId)
+      const awardsWithSize = await prisma.specialAward.findMany({
+        where: { id: { in: awardIds } },
+        select: { id: true, shortlistSize: true },
+      })
+
+      for (const award of awardsWithSize) {
+        const eligible = awardMatchesByAward.get(award.id) || []
+        const shortlistSize = award.shortlistSize ?? 10
+        const topN = eligible
+          .sort((a, b) => b.qualityScore - a.qualityScore)
+          .slice(0, shortlistSize)
+
+        if (topN.length > 0) {
+          await prisma.$transaction(
+            topN.map((e) =>
+              prisma.awardEligibility.update({
+                where: {
+                  awardId_projectId: {
+                    awardId: award.id,
+                    projectId: e.projectId,
+                  },
+                },
+                data: { shortlisted: true },
+              })
+            )
+          )
+        }
+
+        // Mark award eligibility job as completed
+        await prisma.specialAward.update({
+          where: { id: award.id },
+          data: {
+            eligibilityJobStatus: 'COMPLETED',
+            eligibilityJobDone: results.length,
+            eligibilityJobTotal: results.length,
+          },
+        })
+      }
+
+      console.log(`[Filtering] Auto-shortlisted for ${awardsWithSize.length} award(s)`)
+    }
 
     // Count outcomes
     const passedCount = results.filter((r) => r.outcome === 'PASSED').length
@@ -498,6 +626,37 @@ export const filteringRouter = router({
         where: { roundId: input.roundId },
       })
 
+      // Clear award eligibilities for awards linked to this competition
+      const roundForComp = await ctx.prisma.round.findUniqueOrThrow({
+        where: { id: input.roundId },
+        select: { competitionId: true },
+      })
+      if (roundForComp.competitionId) {
+        const linkedAwards = await ctx.prisma.specialAward.findMany({
+          where: {
+            competitionId: roundForComp.competitionId,
+            useAiEligibility: true,
+          },
+          select: { id: true },
+        })
+        const awardIds = linkedAwards.map((a) => a.id)
+        if (awardIds.length > 0) {
+          await ctx.prisma.awardEligibility.deleteMany({
+            where: { awardId: { in: awardIds } },
+          })
+          await ctx.prisma.specialAward.updateMany({
+            where: { id: { in: awardIds } },
+            data: {
+              eligibilityJobStatus: null,
+              eligibilityJobTotal: null,
+              eligibilityJobDone: null,
+              eligibilityJobError: null,
+              eligibilityJobStarted: null,
+            },
+          })
+        }
+      }
+
       const job = await ctx.prisma.filteringJob.create({
         data: {
           roundId: input.roundId,
diff --git a/src/server/services/ai-filtering.ts b/src/server/services/ai-filtering.ts
index 4a2ffb3..7ab667f 100644
--- a/src/server/services/ai-filtering.ts
+++ b/src/server/services/ai-filtering.ts
@@ -98,6 +98,21 @@ export interface ProjectFilteringResult {
   outcome: 'PASSED' | 'FILTERED_OUT' | 'FLAGGED'
   ruleResults: RuleResult[]
   aiScreeningJson?: Record<string, unknown>
+  awardMatches?: AwardMatchResult[]
+}
+
+export type AwardCriteriaInput = {
+  awardId: string
+  awardName: string
+  criteriaText: string
+}
+
+export type AwardMatchResult = {
+  awardId: string
+  eligible: boolean
+  confidence: number
+  qualityScore: number
+  reasoning: string
 }
 
 interface ProjectForFiltering {
@@ -381,6 +396,7 @@ interface AIScreeningResult {
   reasoning: string
   qualityScore: number
   spamRisk: boolean
+  awardMatches?: AwardMatchResult[]
 }
 
 /**
@@ -393,7 +409,8 @@ async function processAIBatch(
   anonymized: AnonymizedProjectForAI[],
   mappings: ProjectAIMapping[],
   userId?: string,
-  entityId?: string
+  entityId?: string,
+  awards?: AwardCriteriaInput[]
 ): Promise<{
   results: Map<string, AIScreeningResult>
   tokensUsed: number
@@ -404,6 +421,34 @@ async function processAIBatch(
   // Sanitize user-supplied criteria
   const { sanitized: safeCriteria } = sanitizeUserInput(criteriaText)
 
+  // Build system prompt — dynamically append awards block if any
+  let systemPrompt = AI_SCREENING_SYSTEM_PROMPT
+  if (awards && awards.length > 0) {
+    const awardsList = awards
+      .map((a, i) => `${i + 1}. ${a.awardName}: ${a.criteriaText}`)
+      .join('\n')
+    systemPrompt += `
+
+## Special Award Evaluation (Additional Task)
+In addition to the main screening, evaluate each project against these special awards.
+Award eligibility is independent of the main screening outcome — a project can fail screening but still match an award.
+
+### Awards
+${awardsList}
+
+Add an "award_matches" array to each project result:
+"award_matches": [
+  {
+    "award_index": 1,
+    "eligible": true/false,
+    "confidence": 0.0-1.0,
+    "quality_score": 0-100,
+    "reasoning": "1-2 sentence explanation"
+  }
+]
+Include one entry per award for every project. quality_score (0-100) measures fit for ranking.`
+  }
+
   // Build user prompt with clear structure
   const userPrompt = `## Screening Criteria
 The admin has defined the following requirements. Evaluate each project against ALL of these criteria:
@@ -418,16 +463,17 @@ Evaluate each project and return JSON with your assessment.`
   const MAX_PARSE_RETRIES = 2
   let parseAttempts = 0
   let response: Awaited<ReturnType<typeof openai.chat.completions.create>>
+  const maxTokens = Math.min(8000, 4000 + (awards?.length ?? 0) * 500)
 
   try {
     const params = buildCompletionParams(model, {
       messages: [
-        { role: 'system', content: AI_SCREENING_SYSTEM_PROMPT },
+        { role: 'system', content: systemPrompt },
         { role: 'user', content: userPrompt },
       ],
       jsonMode: true,
       temperature: 0.1,
-      maxTokens: 4000,
+      maxTokens,
     })
 
     response = await openai.chat.completions.create(params)
@@ -457,6 +503,13 @@ Evaluate each project and return JSON with your assessment.`
         reasoning: string
         quality_score: number
         spam_risk: boolean
+        award_matches?: Array<{
+          award_index: number
+          eligible: boolean
+          confidence: number
+          quality_score: number
+          reasoning: string
+        }>
       }>
     }
 
@@ -476,12 +529,12 @@ Evaluate each project and return JSON with your assessment.`
           // Retry the API call with hint
           const retryParams = buildCompletionParams(model, {
             messages: [
-              { role: 'system', content: AI_SCREENING_SYSTEM_PROMPT },
+              { role: 'system', content: systemPrompt },
               { role: 'user', content: userPrompt + '\n\nIMPORTANT: Please ensure valid JSON output.' },
             ],
             jsonMode: true,
             temperature: 0.1,
-            maxTokens: 4000,
+            maxTokens,
           })
           response = await openai.chat.completions.create(retryParams)
           const retryUsage = extractTokenUsage(response)
@@ -496,12 +549,31 @@ Evaluate each project and return JSON with your assessment.`
     for (const result of parsed.projects || []) {
       const mapping = mappings.find((m) => m.anonymousId === result.project_id)
       if (mapping) {
+        // Map award matches back to real award IDs
+        let awardMatches: AwardMatchResult[] | undefined
+        if (awards && awards.length > 0 && result.award_matches) {
+          awardMatches = []
+          for (const match of result.award_matches) {
+            const award = awards[(match.award_index ?? 1) - 1]
+            if (award) {
+              awardMatches.push({
+                awardId: award.awardId,
+                eligible: match.eligible,
+                confidence: match.confidence,
+                qualityScore: Math.max(0, Math.min(100, match.quality_score ?? 0)),
+                reasoning: match.reasoning ?? '',
+              })
+            }
+          }
+        }
+
         results.set(mapping.realId, {
           meetsCriteria: result.meets_criteria,
           confidence: result.confidence,
           reasoning: result.reasoning,
           qualityScore: result.quality_score,
           spamRisk: result.spam_risk,
+          awardMatches: awardMatches && awardMatches.length > 0 ? awardMatches : undefined,
         })
       }
     }
@@ -550,7 +622,8 @@ export async function executeAIScreening(
   userId?: string,
   entityId?: string,
   onProgress?: ProgressCallback,
-  onBatchComplete?: (batchResults: Map<string, AIScreeningResult>) => Promise<void>
+  onBatchComplete?: (batchResults: Map<string, AIScreeningResult>) => Promise<void>,
+  awards?: AwardCriteriaInput[]
 ): Promise<Map<string, AIScreeningResult>> {
   const results = new Map<string, AIScreeningResult>()
 
@@ -623,7 +696,8 @@ export async function executeAIScreening(
           batch.anon,
           batch.maps,
           userId,
-          entityId
+          entityId,
+          awards
         )
         return { batchResults, tokensUsed, index: batch.index }
       })
@@ -705,7 +779,8 @@ export async function executeFilteringRules(
   userId?: string,
   roundId?: string,
   onProgress?: ProgressCallback,
-  onResultsBatch?: (results: ProjectFilteringResult[]) => Promise<void>
+  onResultsBatch?: (results: ProjectFilteringResult[]) => Promise<void>,
+  awards?: AwardCriteriaInput[]
 ): Promise<ProjectFilteringResult[]> {
   const activeRules = rules
     .filter((r) => r.isActive)
@@ -750,7 +825,8 @@ export async function executeFilteringRules(
   function computeProjectResult(
     projectId: string,
     aiRuleResults: Array<{ ruleId: string; ruleName: string; passed: boolean; action: string; reasoning?: string }>,
-    aiScreeningData: Record<string, unknown>
+    aiScreeningData: Record<string, unknown>,
+    awardMatches?: AwardMatchResult[]
   ): ProjectFilteringResult {
     const nonAi = nonAiEval.get(projectId)!
     const ruleResults: RuleResult[] = [...nonAi.ruleResults]
@@ -777,6 +853,7 @@ export async function executeFilteringRules(
       outcome: hasFailed ? 'FILTERED_OUT' : hasFlagged ? 'FLAGGED' : 'PASSED',
       ruleResults,
       aiScreeningJson: Object.keys(aiScreeningData).length > 0 ? aiScreeningData : undefined,
+      awardMatches: awardMatches && awardMatches.length > 0 ? awardMatches : undefined,
     }
   }
 
@@ -820,13 +897,14 @@ export async function executeFilteringRules(
           computeProjectResult(
             projectId,
             [{ ruleId: aiRule.id, ruleName: aiRule.name, passed, action: aiAction, reasoning: aiResult.reasoning }],
-            { [aiRule.id]: aiResult }
+            { [aiRule.id]: aiResult },
+            aiResult.awardMatches
           )
         )
       }
       allResults.push(...batchResults)
       if (onResultsBatch) await onResultsBatch(batchResults)
-    })
+    }, awards)
 
     return allResults
   }
@@ -835,7 +913,7 @@ export async function executeFilteringRules(
   const aiResults = new Map<string, Map<string, AIScreeningResult>>()
   for (const aiRule of aiRules) {
     const config = aiRule.configJson as unknown as AIScreeningConfig
-    const screeningResults = await executeAIScreening(config, projects, userId, roundId, onProgress)
+    const screeningResults = await executeAIScreening(config, projects, userId, roundId, onProgress, undefined, awards)
     aiResults.set(aiRule.id, screeningResults)
   }
 
@@ -870,7 +948,21 @@ export async function executeFilteringRules(
         aiScreeningData[aiRule.id] = screening
       }
     }
-    results.push(computeProjectResult(project.id, aiRuleResults, aiScreeningData))
+    // Merge award matches from all AI rules (dedup by awardId, first wins)
+    const mergedAwardMatches: AwardMatchResult[] = []
+    const seenAwardIds = new Set<string>()
+    for (const aiRule of aiRules) {
+      const screening = aiResults.get(aiRule.id)?.get(project.id)
+      if (screening?.awardMatches) {
+        for (const m of screening.awardMatches) {
+          if (!seenAwardIds.has(m.awardId)) {
+            seenAwardIds.add(m.awardId)
+            mergedAwardMatches.push(m)
+          }
+        }
+      }
+    }
+    results.push(computeProjectResult(project.id, aiRuleResults, aiScreeningData, mergedAwardMatches))
   }
 
   if (onResultsBatch) await onResultsBatch(results)