■FIG. 00 · Portfolio · Final-year undergraduate · IIT Kanpur
Saubhagya Pandey
I build production software and applied AI systems — payments cores that stay idempotent under retry storms, on-device speech pipelines, LLM routers that know when to abstain. And I keep a second, separate practice: credit-risk scorecards and market-risk analytics, where the discipline is the same but the stakes read differently. Two doors below. Each one stands on its own; pick the one that matches why you're here.
■ DOOR A — THE LARGER PRACTICE
Engineering & AI
Software engineering, product instinct and applied AI read off the same body of work here: a solo-built production SaaS, a Go data plane, an on-device C++ speech pipeline, an abstaining LLM router, a leaderboard-ranked retrieval system, and a community product shipped from survey to launch.
6 PROJECTS · 3 WORK NARRATIVES · LIVE DEMOS WHERE THE ARTIFACTS ALLOW
Enter Engineering & AI →■ DOOR B — THE SECOND PRACTICE
Markets & Risk
Credit-risk scorecards with auditable decision APIs, a reconciliation-grade market-data lake, lender equity research, and document AI for bank statements. A separate craft, held to the same standard: every number reproducible.
4 PROJECTS · DESK ANALYTICS NARRATIVE
Enter Markets & Risk →On breadth
The two practices are deliberately kept separate below this page — different artifacts, different proofs — but they share one habit: claims only survive here if an artifact backs them.
On the demos
Interactive figures are either wired to committed repository artifacts, verbatim ports of deterministic code, or labeled behavioral models. Measured numbers appear only as static charts of real benchmark output.
On the person
Hindi poet with two Inter-IIT gold medals, chess bronze medallist, coordinator of a 33-member campus literary team. The human layer lives in About.