Portfolio · Pune, IN
Hey, I'm Kawas.
Backend developer. Co-founded DNA. Started with classification models and Hugging Face in undergrad — now reading research papers, following where the field is going. The direction is clear: Gen AI → Agentic AI → what comes next. I'm moving deeper.
Status
key takeaway
ownership across every role
CROSS-POLLINATION LOG
— no, no, everything connects.
[RESEARCH YEAR: 2025]
HOW I GOT HERE
DISCORD NATION ALPHA(REF: CREATOR ECONOMY)
Co-founded Gen-Z digital creator and community network. Scaled to 4,300+ members across Discord — video editing, content writing, graphic design, audience building. Worked across multiple hierarchy levels: operations, moderation, engagement strategy, creator collaboration. tried everything at once Achieved ~35% engagement growth via data-driven strategy and creator collaborations.
↳ owned every role — operations, moderation, creator strategy
35% growth
data-driven strategy + creator collabs
FOOD RECIPES BOT (RAG · FAISS · DJANGO)
First end-to-end retrieval pipeline shipped. 80% of RAG quality lives in retrieval, not the model not the generation. Chunking strategy, embedding model choice, metadata filters — chunking strategy clicked here.
→ re-read the RAG paper after building — it reads differently
HOW DOMAINS CROSS-POLLINATE
BUILD. LEARN. ITERATE.
Not "plan endlessly." Not "wait until ready." Build something. Learn from it honestly. Iterate until it works — or until you understand why it cannot. DNA was run this way. The Food Recipes bot was built this way. This site was designed this way.
↳ the iterate part is what everyone skips — real learning lives here
QUEUED & CONCEPTUAL BLENDING
CURRENTLY
“A complex system that works is invariably found to have evolved from a simple system that worked.”
John Gall — Systemantics, 1975
RAG pipeline
↑ 80% quality lives here
What community taught me
Running DNA was a leadership challenge — ownership across moderation, creator collaboration, and data-driven retention for 4,300+ members. Communication and adaptability were the real skills.
Read DDIA Ch.3
Storage & Retrieval. The indexing section finally clicked when building the FAISS pipeline.
Embedding space
token → vector space
Project stack
Memory in AI Agents
How do you give a model persistent context without blowing the window? Same chunking problem as RAG, one abstraction up.
Gen AI → Agentic
The shift from "generate a response" to "complete a task" changes the whole stack. Memory, planning, tool use. That's the direction.
Attention is All You Need
Vaswani et al., 2017. At the time: just another arXiv paper. By 2026: the architecture behind every model you use.