Anas Khan.
Engineering cognitive flows.
AI Systems Engineer specializing in multi-agent workflows, vector retrieval indexes, and high-performance edge inference. Currently pursuing an MSc in Artificial Intelligence at Imperial College London, designing low-latency architectures for cognitive compute.
attention matrix
key-query scaling alignment maps
Attention (24494)Ġis (37732)Ġall (35138)Ġyou (43545)Ġneed (39910)
syntactic / local — focuses on adjacent tokens and immediate local word structures.
temperatureτ = 1.0
attention formula
A(Q,K,V) =softmax(
Q Kᵀτ · √dₖ
)VAttention
is
all
you
need
Attention
85%
14%
0%
0%
0%
is
12%
75%
12%
0%
0%
all
0%
12%
75%
12%
0%
you
0%
0%
12%
75%
12%
need
0%
0%
0%
14%
85%
Y: QUERY (Q)
X: KEY (K)
X: KEY (K)
vector audit
q_idx: --→k_idx: --
RAW DOT PROD:0.000
SCALED ALIGN:0.000
SOFTMAX WEIGHT:0.0%