diff --git a/src/context/analyzer.ts b/src/context/analyzer.ts
index 57f2631..36bc2e2 100644
--- a/src/context/analyzer.ts
+++ b/src/context/analyzer.ts
@@ -69,61 +69,389 @@ const STOP_WORDS = new Set([
 	"each",
 	"every",
 	"both",
+	"can",
+	"could",
+	"may",
+	"might",
+	"must",
+	"shall",
+	"should",
+	"will",
+	"would",
+	"just",
+	"now",
+	"then",
+	"there",
+	"here",
+	"also",
+	"only",
+	"very",
+	"really",
+	"quite",
+	"rather",
+	"some",
+	"any",
+	"many",
+	"much",
 	"few",
 	"more",
 	"most",
-	"other",
-	"some",
+	"less",
+	"least",
+	"than",
 	"such",
+	"other",
+	"another",
+	"same",
+	"different",
+	"new",
+	"old",
+	"good",
+	"bad",
+	"big",
+	"small",
+	"large",
+	"short",
+	"long",
+	"high",
+	"low",
+	"right",
+	"wrong",
+	"true",
+	"false",
+	"yes",
 	"no",
-	"nor",
 	"not",
-	"only",
-	"own",
-	"same",
-	"so",
-	"than",
-	"too",
-	"very",
-	"just",
-	"can",
-	"should",
+	"never",
+	"always",
+	"often",
+	"sometimes",
+	"usually",
+	"generally",
+	"specifically",
+	"particularly",
+	"actually",
+	"really",
+	"certainly",
+	"definitely",
+	"probably",
+	"possibly",
+	"maybe",
+	"perhaps",
+	"like",
+	"such",
+	"well",
+	"even",
+	"though",
+	"although",
+	"because",
+	"since",
+	"while",
+	"during",
+	"after",
+	"before",
+	"until",
+	"since",
+	"ago",
+	"later",
+	"earlier",
+	"soon",
 	"now",
-	"into",
-	"also",
-	"been",
-	"being",
-	"do",
-	"does",
-	"did",
-	"doing",
-	"would",
-	"could",
-	"might",
-	"must",
-	"shall",
-	"may",
-	"add",
-	"added",
-	"adding",
+	"then",
+	"when",
+	"where",
+	"why",
+	"how",
+	"what",
+	"which",
+	"who",
+	"whose",
+	"whom",
+	"that",
+	"this",
+	"these",
+	"those",
+	"i",
+	"me",
+	"my",
+	"myself",
+	"you",
+	"your",
+	"yourself",
+	"he",
+	"him",
+	"his",
+	"himself",
+	"she",
+	"her",
+	"hers",
+	"herself",
+	"it",
+	"its",
+	"itself",
+	"we",
+	"us",
+	"our",
+	"ourselves",
+	"they",
+	"them",
+	"their",
+	"themselves",
+	"one",
+	"ones",
+	"someone",
+	"somebody",
+	"something",
+	"everyone",
+	"everybody",
+	"everything",
+	"nobody",
+	"nothing",
+	"none",
+	"anybody",
+	"anything",
+	"each",
+	"either",
+	"neither",
+	"both",
+	"all",
+	"any",
+	"some",
+	"many",
+	"much",
+	"few",
+	"little",
+	"more",
+	"most",
+	"least",
+	"less",
+	"than",
+	"such",
+	"other",
+	"another",
+	"same",
+	"different",
+	"changing",
+]);
+
+// Generic keywords that don't provide meaningful cluster themes
+const GENERIC_KEYWORDS = new Set([
+	// Source/tool specific
+	"opencode",
+	"background",
+	"zsh",
+	"bash",
+	"terminal",
+	"shell",
+	"command",
+	"commands",
+	"git",
+	"file",
+	"files",
+	"vscode",
+	"cursor",
+	"editor",
+	"code",
+	"coding",
+	"development",
+	"programming",
+	"software",
+
+	// Generic work terms
+	"project",
+	"projects",
+	"work",
+	"working",
+	"task",
+	"tasks",
+	"item",
+	"items",
+	"activity",
+	"activities",
+	"session",
+	"sessions",
+	"window",
+	"windows",
+	"tab",
+	"tabs",
+	"page",
+	"pages",
+
+	// Generic data/tech terms
+	"data",
+	"database",
+	"db",
+	"api",
+	"apis",
+	"server",
+	"servers",
+	"client",
+	"clients",
+	"request",
+	"requests",
+	"response",
+	"responses",
+	"error",
+	"errors",
+
+	// Generic issue/PR terms
+	"issue",
+	"issues",
+	"bug",
+	"bugs",
 	"fix",
-	"fixed",
-	"fixing",
+	"fixes",
+	"feature",
+	"features",
 	"update",
-	"updated",
-	"updating",
-	"implement",
-	"implemented",
-	"implementing",
+	"updates",
+	"change",
+	"changes",
+	"commit",
+	"commits",
+	"push",
+	"pull",
+	"merge",
+	"branch",
+	"branches",
+	"repository",
+	"repositories",
+	"repo",
+	"repos",
+
+	// GitHub specific
+	"github",
+	"pr",
+	"pull",
+	"request",
+	"opened",
+	"merged",
+	"closed",
+
+	// User/repo names (these are too generic in PR titles)
+	"jvalentini",
+	"worklog",
+	"main",
+	"master",
+	"develop",
+	"development",
+
+	// Common verbs that don't help clustering
+	"add",
+	"added",
 	"create",
 	"created",
-	"creating",
+	"update",
+	"updated",
 	"remove",
 	"removed",
-	"removing",
-	"change",
-	"changed",
-	"changing",
+	"delete",
+	"deleted",
+	"implement",
+	"implemented",
+	"improve",
+	"improved",
+	"fix",
+	"fixed",
+	"ensure",
+	"ensures",
+	"make",
+	"made",
+	"set",
+	"setting",
+	"get",
+	"getting",
+	"run",
+	"running",
+	"start",
+	"started",
+	"stop",
+	"stopped",
+	"enable",
+	"enabled",
+	"disable",
+	"disabled",
+
+	// PR/commit specific terms
+	"pr",
+	"pull",
+	"request",
+	"requests",
+	"merge",
+	"merges",
+	"merged",
+	"commit",
+	"commits",
+	"branch",
+	"branches",
+
+	// More generic action words
+	"trigger",
+	"triggers",
+	"triggered",
+	"handle",
+	"handles",
+	"handled",
+	"support",
+	"supports",
+	"supported",
+	"resolve",
+	"resolves",
+	"resolved",
+	"address",
+	"addresses",
+	"addressed",
+
+	// Tool/framework names that are too generic
+	"release",
+	"please",
+	"npm",
+	"bun",
+	"node",
+	"typescript",
+	"ts",
+	"js",
+	"javascript",
+	"react",
+	"vue",
+	"angular",
+	"next",
+	"nuxt",
+	"express",
+	"fastify",
+	"docker",
+	"kubernetes",
+	"aws",
+	"gcp",
+	"azure",
+	"postgres",
+	"postgresql",
+	"mysql",
+	"mongodb",
+	"redis",
+	"elasticsearch",
+
+	// Even more generic terms
+	"new",
+	"old",
+	"good",
+	"bad",
+	"better",
+	"best",
+	"simple",
+	"complex",
+	"easy",
+	"hard",
+	"quick",
+	"fast",
+	"slow",
+	"small",
+	"large",
+	"big",
+	"little",
+	"major",
+	"minor",
+	"patch",
 ]);
 
 function tokenize(text: string): string[] {
@@ -223,7 +551,28 @@ export function computeSimilarityMatrix(items: WorkItem[]): number[][] {
 			} else {
 				const vecJ = vectors[j];
 				if (vecJ) {
-					row[j] = cosineSimilarity(vecI, vecJ);
+					let similarity = cosineSimilarity(vecI, vecJ);
+
+					// Boost similarity for clearly related items
+					const itemI = items[i];
+					const itemJ = items[j];
+					if (itemI && itemJ) {
+						// GitHub PR events with same PR number should be highly similar
+						const prMatchI = itemI.title.match(/PR #(\d+)/);
+						const prMatchJ = itemJ.title.match(/PR #(\d+)/);
+						if (prMatchI && prMatchJ && prMatchI[1] === prMatchJ[1]) {
+							similarity = Math.max(similarity, 0.9);
+						}
+
+						// Same commit hash should be highly similar
+						const commitMatchI = itemI.title.match(/([a-f0-9]{7,40})/);
+						const commitMatchJ = itemJ.title.match(/([a-f0-9]{7,40})/);
+						if (commitMatchI && commitMatchJ && commitMatchI[1] === commitMatchJ[1]) {
+							similarity = Math.max(similarity, 0.95);
+						}
+					}
+
+					row[j] = similarity;
 				}
 			}
 		}
@@ -313,13 +662,75 @@ function computeClusterCoherence(items: WorkItem[]): number {
 }
 
 function generateThemeLabel(keywords: string[], items: WorkItem[]): string {
-	if (keywords.length === 0) {
+	// Filter out generic keywords that don't provide meaningful themes
+	const meaningfulKeywords = keywords.filter((k) => !GENERIC_KEYWORDS.has(k.toLowerCase()));
+
+	if (meaningfulKeywords.length === 0) {
+		// If no meaningful keywords, try to generate a theme from item content
+		const allText = items.map((i) => `${i.title} ${i.description ?? ""}`).join(" ");
+		const titleTokens = tokenize(allText).filter((t) => !GENERIC_KEYWORDS.has(t.toLowerCase()));
+
+		if (titleTokens.length > 0) {
+			// Use the most common meaningful words from all content
+			const tokenCounts = new Map<string, number>();
+			for (const token of titleTokens) {
+				tokenCounts.set(token, (tokenCounts.get(token) ?? 0) + 1);
+			}
+			const sortedTokens = [...tokenCounts.entries()].sort((a, b) => b[1] - a[1]).slice(0, 3); // Get top 3 most meaningful terms
+
+			// Try to find the most specific/unique terms
+			const meaningfulTokens = sortedTokens
+				.map(([token]) => token)
+				.filter((token) => token.length > 3) // Prefer longer, more specific words
+				.slice(0, 2); // Take at most 2
+
+			if (meaningfulTokens.length >= 2) {
+				const first = meaningfulTokens[0];
+				const second = meaningfulTokens[1];
+				if (first && second) {
+					return `${capitalize(first)} & ${capitalize(second)}`;
+				}
+			} else if (meaningfulTokens.length === 1) {
+				const first = meaningfulTokens[0];
+				if (first) {
+					return capitalize(first);
+				}
+			}
+		}
+
+		// Last resort: try to find any nouns or proper names
+		const fullContent = items.map((i) => `${i.title} ${i.description ?? ""}`).join(" ");
+		const matchedWords = fullContent.toLowerCase().match(/\b[a-z]{4,}\b/g); // Words of 4+ characters
+		const fallbackTokens: string[] =
+			matchedWords?.filter((word): word is string => !GENERIC_KEYWORDS.has(word))?.slice(0, 2) ??
+			[];
+
+		if (fallbackTokens.length >= 2) {
+			const first = fallbackTokens[0];
+			const second = fallbackTokens[1];
+			if (first && second) {
+				return `${capitalize(first)} & ${capitalize(second)}`;
+			}
+		} else if (fallbackTokens.length === 1) {
+			const first = fallbackTokens[0];
+			if (first) {
+				return capitalize(first);
+			}
+		}
+
+		// Ultimate fallback to source-based naming
 		const sources = [...new Set(items.map((i) => i.source))];
 		return sources.length === 1 ? `${sources[0]} activity` : "Mixed activity";
 	}
 
-	const primary = keywords[0];
-	const secondary = keywords[1];
+	// For meaningful keywords, prefer longer, more specific ones
+	const sortedKeywords = meaningfulKeywords
+		.sort((a, b) => b.length - a.length) // Longer words first
+		.slice(0, 2);
+
+	const primary = sortedKeywords[0];
+	const secondary = sortedKeywords[1];
+
 	if (!primary) {
 		const sources = [...new Set(items.map((i) => i.source))];
 		return sources.length === 1 ? `${sources[0]} activity` : "Mixed activity";
@@ -347,9 +758,14 @@ export function findCrossClusterConnections(
 			const b = clusters[j];
 			if (!a || !b) continue;
 
-			const sharedKeywords = a.keywords.filter((k) => b.keywords.includes(k));
+			// Only consider meaningful shared keywords (filter out generic ones)
+			const sharedKeywords = a.keywords
+				.filter((k) => b.keywords.includes(k))
+				.filter((k) => !GENERIC_KEYWORDS.has(k.toLowerCase()));
 
-			if (sharedKeywords.length > 0) {
+			// Only create connections if there are meaningful shared keywords
+			// and the clusters are actually different (not just the same generic keywords)
+			if (sharedKeywords.length > 0 && a.theme !== b.theme) {
 				connections.push({
 					from: a.id,
 					to: b.id,