01 / CODE INTELLIGENCE
Sydes
AI system understanding for real codebases. Trace API flows across services, identify downstream calls, detect sinks, and export system graphs. Generate risks and integration tests
AI systems · inference optimization · code intelligence
Karimulla Naik
Independent builder working on AI system understanding, LLM inference experiments, and alternative sequence architectures. I like tools that make complex software and models inspectable.
$ sydes trace "/users" --method POST → matched endpoint repo: api handler: create_user → traced downstream calls route → service → repository → detected sink database write: db.commit() artifact: trace_graph.json
Featured projects
━━━━━━━━ SYSTEMS / OSS
02 / INCIDENT ANALYSIS
AI Autopsy
CLI-first incident and log analysis experiments using local and remote LLMs to interpret events, timelines, and root-cause candidates. It makes it easy to understand multi-log timelines.
03 / MULTI AGENT SYSTEMS
Bio Agents
Experimental autonomous agent workflows using bio-inspired algorithms, orchestration, and AI-assisted scientific exploration. It works by evolving the best answer through competition.
Research taste
━━━━━━━━ EXPERIMENTS
I am interested in systems where the boundary between software engineering, ML research, and infrastructure becomes blurry.
Program understanding
Turning real codebases into inspectable flow graphs.
Inference optimization
Memory-aware serving, KV-cache design, and decode-time shortcuts.
Sequence architecture
Exploring spectral, recurrent, and low-rank alternatives to attention.
$ python train_assm.py model: spectral-shift mixer goal: reduce attention dependence eval: val_bpb, throughput, params step 0200 val_bpb=1.31 step 0400 val_bpb=1.28 step 0600 val_bpb=1.26 notes: keep experiments small, measurable, reproducible
Writing & notebooks
━━━━━━━━ NOTES