diff --git a/integration/grpc_test.go b/integration/grpc_test.go
index 3eb972e..bc0d8bc 100644
--- a/integration/grpc_test.go
+++ b/integration/grpc_test.go
@@ -3,7 +3,6 @@ package integration_test
 import (
 	"context"
 	"fmt"
-	"log"
 	"os"
 	"testing"
 
@@ -96,12 +95,12 @@ func deleteEverything(t *testing.T, rc *rockset.RockClient, cfg rss.Config) func
 
 			for _, d := range response.GetData() {
 				if d.GetStatus() != "DELETED" {
-					log.Printf("failed to delete %s: %s", d.GetId(), d.GetStatus())
+					t.Logf("failed to delete %s: %s", d.GetId(), d.GetStatus())
 				}
 			}
 			t.Logf("%s: deleted %d documents", coll, len(response.GetData()))
 
-			// wait for the delete to complete
+			// wait for the deletion to have propagated
 			w := wait.New(rc)
 			if err = w.UntilQueryable(ctx, cfg.Workspace, coll, []string{response.GetLastOffset()}); err != nil {
 				return err
diff --git a/storage/spanstore/store.go b/storage/spanstore/store.go
index 97ae19a..0284e09 100644
--- a/storage/spanstore/store.go
+++ b/storage/spanstore/store.go
@@ -204,13 +204,8 @@ func (s Store) findTraces(ctx context.Context, ids []model.TraceID) ([]*model.Tr
 
 // TODO: this is very naïve and must be improved to avoid SQL injection
 func buildQuery(config Config, query *spanstore.TraceQueryParameters) string {
-	// TODO implement tag filtering
-	// {
-	// 	Tags:map[sameplacetag1:sameplacevalue sameplacetag2:123 sameplacetag3:72.5 sameplacetag4:true]
-	// }
-
 	var q strings.Builder
-	q.WriteString("SELECT trace_id, MIN(start_time) AS start_time\n")
+	q.WriteString("SELECT spans.trace_id AS trace_id, MIN(spans.start_time) AS start_time\n")
 	q.WriteString(fmt.Sprintf("FROM %s.%s spans\n", config.Workspace, config.Spans))
 
 	q.WriteString("WHERE ")
@@ -239,6 +234,10 @@ func buildQuery(config Config, query *spanstore.TraceQueryParameters) string {
 		q.WriteString(fmt.Sprintf(" AND spans.duration <= %d", query.DurationMax))
 	}
 
+	for k, v := range query.Tags {
+		q.WriteString(fmt.Sprintf(" AND spans.kv.%s = '%s'", k, v))
+	}
+
 	q.WriteString("\nGROUP BY trace_id\n")
 	q.WriteString("ORDER BY start_time DESC, trace_id\n")
 
@@ -286,13 +285,3 @@ func traceIDs(ids []model.TraceID) (string, error) {
 
 	return strings.Join(tids, ","), nil
 }
-
-/*
-SELECT trace_id, MIN(start_time) AS start_time
-FROM test.spans spans
-WHERE spans.process.service_name = 'query11-service'
-AND spans.start_time \u003e= '2024-02-11T15:46:31.639875Z' AND
-spans.start_time \u003c= '2024-02-11T17:46:31.639875Z'\n
-GROUP BY trace_id\nORDER BY start_time DESC, trace_id\n
-LIMIT 1000
-*/
diff --git a/storage/spanstore/writer.go b/storage/spanstore/writer.go
index 953576f..3fc2644 100644
--- a/storage/spanstore/writer.go
+++ b/storage/spanstore/writer.go
@@ -2,6 +2,7 @@ package spanstore
 
 import (
 	"context"
+	"strconv"
 
 	"github.com/jaegertracing/jaeger/model"
 	"github.com/rockset/rockset-go-client/writer"
@@ -9,13 +10,55 @@ import (
 
 const unspecified = "unspecified"
 
+type Span struct {
+	model.Span
+	KV map[string]string `json:"kv"`
+}
+
+func extractKeyAndValue(tag model.KeyValue) (k, v string) {
+	var s string
+
+	switch tag.VType {
+	case model.StringType:
+		s = tag.VStr
+	case model.BoolType:
+		s = strconv.FormatBool(tag.VBool)
+	case model.Int64Type:
+		s = strconv.FormatInt(tag.VInt64, 10)
+	case model.Float64Type:
+		s = strconv.FormatFloat(tag.VFloat64, 'f', -1, 64)
+	case model.BinaryType:
+		s = string(tag.VBinary)
+	}
+
+	return tag.Key, s
+}
+
 func (s Store) WriteSpan(_ context.Context, span *model.Span) error {
+	// to speed up queries we convert tags & process tags to a single map of string keys and string values,
+	// as that is what we get from the web ui when someone is searching for a trace,
+	// which makes the query much faster as we index the keys and values.
+	sp := Span{
+		Span: *span,
+		KV:   make(map[string]string),
+	}
+
+	for _, tag := range span.Tags {
+		k, v := extractKeyAndValue(tag)
+		sp.KV[k] = v
+	}
+	for _, tag := range span.Process.Tags {
+		k, v := extractKeyAndValue(tag)
+		sp.KV[k] = v
+	}
+
 	s.writer.C() <- writer.Request{
 		Workspace:  s.config.Workspace,
 		Collection: s.config.Spans,
-		Data:       *span,
+		Data:       sp,
 	}
 
+	// TODO we should batch these updates, to reduce the number of writes
 	kind := unspecified
 	for _, tag := range span.Tags {
 		if tag.Key == "span.kind" {