Samsung R&D & Micron: AI / ML Engineer

def min_path(grid):
    R, C = len(grid), len(grid[0]); dp = [0] * C
    for r in range(R):
        for c in range(C):
            if r == 0 and c == 0: dp[c] = grid[0][0]
            elif r == 0: dp[c] = dp[c-1] + grid[r][c]
            elif c == 0: dp[c] = dp[c] + grid[r][c]
            else: dp[c] = min(dp[c], dp[c-1]) + grid[r][c]
    return dp[-1]

• O(R·C) time and O(C) space with a rolling array. For four-direction moves with non-negative weights, switch to Dijkstra.

def islands(g):
    R, C = len(g), len(g[0]); seen = set(); n = 0
    for r in range(R):
        for c in range(C):
            if g[r][c] == 1 and (r, c) not in seen:
                n += 1; stack = [(r, c)]; seen.add((r, c))
                while stack:
                    x, y = stack.pop()
                    for dx, dy in ((1,0),(-1,0),(0,1),(0,-1)):
                        a, b = x + dx, y + dy
                        if 0 <= a < R and 0 <= b < C and g[a][b] == 1 and (a, b) not in seen:
                            seen.add((a, b)); stack.append((a, b))
    return n

1. DNS resolution (browser cache → OS → resolver → root/TLD/authoritative servers) gives the IP address.
2. TCP three-way handshake (SYN, SYN-ACK, ACK), then the TLS handshake for HTTPS (certificate check, key exchange).
3. HTTP request → server (load balancer → web server/app) → response.
4. The browser parses HTML, fetches CSS/JS/images (more requests, often HTTP/2 multiplexed), and renders. - Know the OSI/TCP-IP layers, TCP vs UDP, ports, and what a subnet mask is.

• TCP: reliable, ordered, connection-based (handshake, retransmission, congestion control). Used for APIs, model serving (HTTP/gRPC) and database connections.
• UDP: connectionless with no delivery guarantee, lower latency. Used for real-time video and audio, telemetry streams, DNS, QUIC (HTTP/3 builds reliability on top of UDP).
• Example: factory sensor telemetry might use UDP or MQTT for speed; model inference APIs use TCP.

• Data: wafer maps and SEM images, process parameters (temperature, pressure, tool ID), test results.
• Vision: CNN/ViT classification of defect patterns (center, edge-ring, scratch, random); segmentation to locate defects. Handle class imbalance and rare new defect types (anomaly detection, few-shot learning).
• Tabular: gradient boosting on process parameters to predict yield; SHAP to find root-cause parameters.
• Ops: the model runs at the edge near the tools, engineers give feedback, and the model is retrained as processes change (drift).

• Strength with evidence: "I debug systematically. I traced a 6% accuracy drop to a sensor calibration change by comparing feature distributions across shifts."
• A real weakness with a fix: "I used to over-polish before sharing. Now I share a rough version after day two to get feedback early."
• Keep it honest, specific and short.