diff --git a/pkg/batch/batch.go b/pkg/batch/batch.go
new file mode 100644
index 0000000..911a7a7
--- /dev/null
+++ b/pkg/batch/batch.go
@@ -0,0 +1,263 @@
+// Package batch provides async batch processing for large deduplication workloads.
+// Jobs are queued in-memory, processed by a background worker pool, and results
+// are retained for a configurable TTL.
+package batch
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"sync"
+	"time"
+
+	"github.com/Siddhant-K-code/distill/pkg/pipeline"
+	"github.com/Siddhant-K-code/distill/pkg/types"
+)
+
+// Status represents the lifecycle state of a batch job.
+type Status string
+
+const (
+	StatusQueued     Status = "queued"
+	StatusProcessing Status = "processing"
+	StatusCompleted  Status = "completed"
+	StatusFailed     Status = "failed"
+)
+
+// Job holds the input, state, and result of a single batch job.
+type Job struct {
+	ID          string
+	Status      Status
+	Chunks      []types.Chunk
+	Options     pipeline.Options
+	Result      []types.Chunk
+	Stats       pipeline.Stats
+	Error       string
+	CreatedAt   time.Time
+	StartedAt   time.Time
+	CompletedAt time.Time
+	Progress    float64 // 0–1
+}
+
+// SubmitRequest is the input for submitting a new batch job.
+type SubmitRequest struct {
+	Chunks  []types.Chunk
+	Options pipeline.Options
+}
+
+// ErrJobNotFound is returned when a job ID does not exist.
+var ErrJobNotFound = errors.New("job not found")
+
+// ErrResultExpired is returned when a job's result has been evicted.
+var ErrResultExpired = errors.New("job result has expired")
+
+// Processor manages the job queue, worker pool, and result store.
+type Processor struct {
+	mu         sync.RWMutex
+	jobs       map[string]*Job
+	queue      chan string
+	resultTTL  time.Duration
+	runner     *pipeline.Runner
+	wg         sync.WaitGroup
+	cancelFunc context.CancelFunc
+}
+
+// Config controls processor behaviour.
+type Config struct {
+	// Workers is the number of concurrent processing goroutines. Default: 4.
+	Workers int
+	// QueueSize is the maximum number of queued jobs. Default: 1000.
+	QueueSize int
+	// ResultTTL is how long completed job results are retained. Default: 24h.
+	ResultTTL time.Duration
+}
+
+// DefaultConfig returns sensible defaults.
+func DefaultConfig() Config {
+	return Config{
+		Workers:   4,
+		QueueSize: 1000,
+		ResultTTL: 24 * time.Hour,
+	}
+}
+
+// NewProcessor creates and starts a Processor with the given config.
+func NewProcessor(cfg Config) *Processor {
+	if cfg.Workers < 0 {
+		cfg.Workers = 4
+	}
+	if cfg.QueueSize <= 0 {
+		cfg.QueueSize = 1000
+	}
+	if cfg.ResultTTL <= 0 {
+		cfg.ResultTTL = 24 * time.Hour
+	}
+
+	ctx, cancel := context.WithCancel(context.Background())
+	p := &Processor{
+		jobs:       make(map[string]*Job),
+		queue:      make(chan string, cfg.QueueSize),
+		resultTTL:  cfg.ResultTTL,
+		runner:     pipeline.New(),
+		cancelFunc: cancel,
+	}
+
+	for i := 0; i < cfg.Workers; i++ {
+		p.wg.Add(1)
+		go p.worker(ctx)
+	}
+
+	go p.evictLoop(ctx)
+
+	return p
+}
+
+// Submit enqueues a new batch job and returns its ID.
+func (p *Processor) Submit(req SubmitRequest) (*Job, error) {
+	id := generateID()
+	job := &Job{
+		ID:        id,
+		Status:    StatusQueued,
+		Chunks:    req.Chunks,
+		Options:   req.Options,
+		CreatedAt: time.Now(),
+	}
+
+	p.mu.Lock()
+	p.jobs[id] = job
+	p.mu.Unlock()
+
+	select {
+	case p.queue <- id:
+	default:
+		p.mu.Lock()
+		delete(p.jobs, id)
+		p.mu.Unlock()
+		return nil, fmt.Errorf("job queue is full")
+	}
+
+	return job, nil
+}
+
+// Get returns the current state of a job.
+func (p *Processor) Get(id string) (*Job, error) {
+	p.mu.RLock()
+	defer p.mu.RUnlock()
+	job, ok := p.jobs[id]
+	if !ok {
+		return nil, ErrJobNotFound
+	}
+	// Return a copy to avoid data races.
+	cp := *job
+	return &cp, nil
+}
+
+// Results returns the deduplicated chunks for a completed job.
+func (p *Processor) Results(id string) ([]types.Chunk, pipeline.Stats, error) {
+	p.mu.RLock()
+	defer p.mu.RUnlock()
+	job, ok := p.jobs[id]
+	if !ok {
+		return nil, pipeline.Stats{}, ErrJobNotFound
+	}
+	if job.Status != StatusCompleted {
+		return nil, pipeline.Stats{}, fmt.Errorf("job %s is %s, not completed", id, job.Status)
+	}
+	return job.Result, job.Stats, nil
+}
+
+// List returns all jobs, optionally filtered by status ("" = all).
+func (p *Processor) List(status Status) []*Job {
+	p.mu.RLock()
+	defer p.mu.RUnlock()
+	var out []*Job
+	for _, j := range p.jobs {
+		if status == "" || j.Status == status {
+			cp := *j
+			out = append(out, &cp)
+		}
+	}
+	return out
+}
+
+// Stop gracefully shuts down the processor, waiting for in-flight jobs.
+func (p *Processor) Stop() {
+	p.cancelFunc()
+	close(p.queue)
+	p.wg.Wait()
+}
+
+// worker processes jobs from the queue.
+func (p *Processor) worker(ctx context.Context) {
+	defer p.wg.Done()
+	for id := range p.queue {
+		if ctx.Err() != nil {
+			return
+		}
+		p.process(ctx, id)
+	}
+}
+
+// process runs the pipeline for a single job.
+func (p *Processor) process(ctx context.Context, id string) {
+	p.mu.Lock()
+	job, ok := p.jobs[id]
+	if !ok {
+		p.mu.Unlock()
+		return
+	}
+	job.Status = StatusProcessing
+	job.StartedAt = time.Now()
+	job.Progress = 0.0
+	p.mu.Unlock()
+
+	result, stats, err := p.runner.Run(ctx, job.Chunks, job.Options)
+
+	p.mu.Lock()
+	defer p.mu.Unlock()
+	job, ok = p.jobs[id]
+	if !ok {
+		return
+	}
+	job.CompletedAt = time.Now()
+	job.Progress = 1.0
+	if err != nil {
+		job.Status = StatusFailed
+		job.Error = err.Error()
+	} else {
+		job.Status = StatusCompleted
+		job.Result = result
+		job.Stats = stats
+	}
+}
+
+// evictLoop removes completed/failed jobs whose results have expired.
+func (p *Processor) evictLoop(ctx context.Context) {
+	ticker := time.NewTicker(5 * time.Minute)
+	defer ticker.Stop()
+	for {
+		select {
+		case <-ctx.Done():
+			return
+		case <-ticker.C:
+			p.evict()
+		}
+	}
+}
+
+func (p *Processor) evict() {
+	cutoff := time.Now().Add(-p.resultTTL)
+	p.mu.Lock()
+	defer p.mu.Unlock()
+	for id, job := range p.jobs {
+		if (job.Status == StatusCompleted || job.Status == StatusFailed) &&
+			job.CompletedAt.Before(cutoff) {
+			delete(p.jobs, id)
+		}
+	}
+}
+
+// generateID returns a simple time-based unique ID.
+func generateID() string {
+	return fmt.Sprintf("batch_%d", time.Now().UnixNano())
+}
diff --git a/pkg/batch/batch_test.go b/pkg/batch/batch_test.go
new file mode 100644
index 0000000..312f3eb
--- /dev/null
+++ b/pkg/batch/batch_test.go
@@ -0,0 +1,134 @@
+package batch
+
+import (
+	"fmt"
+	"testing"
+	"time"
+
+	"github.com/Siddhant-K-code/distill/pkg/pipeline"
+	"github.com/Siddhant-K-code/distill/pkg/types"
+)
+
+func makeChunks(n int) []types.Chunk {
+	chunks := make([]types.Chunk, n)
+	for i := range chunks {
+		chunks[i] = types.Chunk{
+			ID:   fmt.Sprintf("chunk-%d", i),
+			Text: "test chunk content for batch processing",
+		}
+	}
+	return chunks
+}
+
+func TestSubmitAndGet(t *testing.T) {
+	p := NewProcessor(Config{Workers: 1, QueueSize: 10, ResultTTL: time.Minute})
+	defer p.Stop()
+
+	job, err := p.Submit(SubmitRequest{
+		Chunks:  []types.Chunk{{ID: "a", Text: "hello"}},
+		Options: pipeline.Options{},
+	})
+	if err != nil {
+		t.Fatalf("Submit: %v", err)
+	}
+	if job.ID == "" {
+		t.Error("expected non-empty job ID")
+	}
+
+	// Poll until done (max 2s).
+	deadline := time.Now().Add(2 * time.Second)
+	for time.Now().Before(deadline) {
+		got, err := p.Get(job.ID)
+		if err != nil {
+			t.Fatalf("Get: %v", err)
+		}
+		if got.Status == StatusCompleted || got.Status == StatusFailed {
+			break
+		}
+		time.Sleep(10 * time.Millisecond)
+	}
+
+	got, _ := p.Get(job.ID)
+	if got.Status != StatusCompleted {
+		t.Errorf("expected StatusCompleted, got %s (error: %s)", got.Status, got.Error)
+	}
+}
+
+func TestResults_NotCompleted(t *testing.T) {
+	p := NewProcessor(Config{Workers: 0, QueueSize: 10, ResultTTL: time.Minute})
+	// Workers=0 means nothing processes; job stays queued.
+	defer p.Stop()
+
+	job, _ := p.Submit(SubmitRequest{
+		Chunks:  []types.Chunk{{ID: "a", Text: "hello"}},
+		Options: pipeline.Options{},
+	})
+
+	_, _, err := p.Results(job.ID)
+	if err == nil {
+		t.Error("expected error for non-completed job")
+	}
+}
+
+func TestGet_NotFound(t *testing.T) {
+	p := NewProcessor(DefaultConfig())
+	defer p.Stop()
+
+	_, err := p.Get("nonexistent")
+	if err != ErrJobNotFound {
+		t.Errorf("expected ErrJobNotFound, got %v", err)
+	}
+}
+
+func TestList(t *testing.T) {
+	p := NewProcessor(Config{Workers: 1, QueueSize: 10, ResultTTL: time.Minute})
+	defer p.Stop()
+
+	p.Submit(SubmitRequest{Chunks: []types.Chunk{{ID: "a", Text: "hello"}}, Options: pipeline.Options{}})
+	p.Submit(SubmitRequest{Chunks: []types.Chunk{{ID: "b", Text: "world"}}, Options: pipeline.Options{}})
+
+	// Wait briefly for processing.
+	time.Sleep(200 * time.Millisecond)
+
+	all := p.List("")
+	if len(all) < 1 {
+		t.Error("expected at least one job in list")
+	}
+}
+
+func TestQueueFull(t *testing.T) {
+	// QueueSize=0 means the channel has no buffer — submit should fail.
+	p := NewProcessor(Config{Workers: 0, QueueSize: 1, ResultTTL: time.Minute})
+	defer p.Stop()
+
+	// Fill the queue.
+	p.Submit(SubmitRequest{Chunks: []types.Chunk{{ID: "a", Text: "x"}}, Options: pipeline.Options{}})
+
+	// This should fail.
+	_, err := p.Submit(SubmitRequest{Chunks: []types.Chunk{{ID: "b", Text: "y"}}, Options: pipeline.Options{}})
+	if err == nil {
+		t.Error("expected error when queue is full")
+	}
+}
+
+func TestDefaultConfig(t *testing.T) {
+	cfg := DefaultConfig()
+	if cfg.Workers <= 0 {
+		t.Error("expected positive workers")
+	}
+	if cfg.ResultTTL <= 0 {
+		t.Error("expected positive ResultTTL")
+	}
+}
+
+func TestGenerateID_Unique(t *testing.T) {
+	ids := map[string]bool{}
+	for i := 0; i < 100; i++ {
+		id := generateID()
+		if ids[id] {
+			t.Errorf("duplicate ID generated: %s", id)
+		}
+		ids[id] = true
+		time.Sleep(time.Nanosecond)
+	}
+}