Shreyansh Kumar Singh
AI & Full-Stack Engineer

For nearly four years at RamanByte he has built the learning platform Classroom+ and client systems on top of it. That work runs for real institutions: a peer-reviewed journal portal, admin, student and faculty apps, and a two-app artisan marketplace launched on a fixed public date.

Alongside it he has built nine independent projects across nine domains: retrieval-augmented generation, autonomous security testing, AI quality assurance, 3D, physics simulation, editorial products, enterprise architecture and scroll-driven motion. The production work taught him what fails when real users arrive. The independent work is where he applies that lesson to newer problems.

He starts with purpose. Before any code, he asks why the thing needs to exist and who is trying to get something done with it. A feature without a reason, in his view, is only another button.

He treats complexity as a chain to be made legible: interface, API, logic, database, model, and the real world it touches. His job is to make that chain feel like one simple experience to the person using it.

He assumes failure. What happens if the API dies, if a user taps twice, if traffic jumps a hundredfold overnight? For the Dada Udyogini launch this was concrete: he designed DadaLoad, a distributed k6 load-testing system run across ten machines, so the team could prove capacity before launch day instead of hoping for it.

He treats AI as infrastructure rather than a feature: the model sits inside the system's loop from the first design, and its output is reviewed like any engineer's code.

• Retrieval-augmented generation. In Sarthi, an AI relationship-manager copilot for banking, he split the pipeline into four modules — embeddings, vector storage, retrieval and generation — so each can be inspected, tested and replaced on its own, and one weak stage can't hide inside the others.
• Agent systems. Nythera runs an LLM agent loop through a seven-phase penetration-testing workflow and confirms each finding with evidence before reporting it. HallogenAI coordinates eight specialized agents to verify bug fixes against live application behavior.
• Grounded products. BookVerse AI generates summaries, timelines, mind maps and a tutor grounded in the source text, across three model paths including local Ollama — and keeps running with no API key.
• Model fluency. He works with Groq, Gemini and Ollama, chooses between hosted and local models by cost and reliability, and designs for the model being unavailable.
• Responsible design. Dangerous capabilities default to off — Nythera's internal-network scanning needs explicit authorization. Data provenance is enforced — Antarang uses only verified Wikimedia sources and refuses to fabricate copyrighted artworks.

Almost every repository ships with a project specification and a handoff document, written so a stranger could pick the work up cold. Visual design is written as a token system before any component exists. Every number in his simulators traces to a published model.

After shipping, he audits his own work. His Sarthi audit ranks findings from P0 to P3 and leads with the worst: no authentication on any endpoint and guessable record IDs. He published it rather than quietly patching it — which says more about his standards than any list of skills.

He is direct. He would rather hear the real problem than sit through a meeting about nothing, and he asks three questions of every new piece of work: what is actually broken, what has been tried, and what done looks like.

He is self-critical without being slow. He has shipped nine projects in five languages while still writing the specification and the audit for each. Speed, in his view, makes rigor affordable — not optional.

He works across roles naturally: specification writer, designer, security reviewer and systems architect, often in the same week on the same project — which he sees as the modern engineer's job rather than a collection of side skills.

What drew him to engineering is the moment a confusing system becomes legible. It's why so much of his independent work starts from a subject he wanted to understand rather than a brief: Indian art history placed properly on the world timeline in Antarang, the physics of blast waves and orbits in The Evolution, the culture of collecting in The Collector's Pulse.

Work where AI is load-bearing and correctness matters: document understanding, agent systems, security, and products that must survive real users. He is open to full-time or hybrid roles in applied AI engineering, AI-focused full-stack engineering, security tooling, and senior .NET and Angular work.