Chemical Engineering undergraduate at IIT Guwahati building at the intersection of product, data, applied AI and quantitative systems.
My work spans product management, consulting, analytics and engineering. GitHub is where I keep the technical side of that work: reproducible prototypes, decision engines, experiments and end-to-end products.
| Area | What interests me |
|---|---|
| Applied AI | Multimodal document understanding, adaptive systems, explainable recommendations and human-in-the-loop decision support |
| Data & decision systems | Turning noisy evidence into auditable rules, models, metrics and operating decisions |
| Full-stack products | Building the complete path from interface and APIs to authentication, databases, deployment and monitoring |
| Quantitative research | Signals, simulation, backtesting, uncertainty and explicit risk controls |
| Project | What it explores | Core tools |
|---|---|---|
| AegisLoop | An adaptive payment-threat digital twin where a contextual bandit searches for evasive campaigns and the defender learns from them | Python, LinUCB, graph features, Next.js |
| TriageLoop AI | Deadline-aware emergency-department decision support that tests clinical need against live queue capacity while keeping clinicians in control | Python, simulation, uncertainty, Next.js |
| Intelligent Invoice Extraction | Multimodal extraction of multilingual invoice fields, signatures and stamps with separate text and visual pipelines | Qwen2.5-VL, QLoRA, RT-DETR, PyTorch |
| ValueLoop AI | A synthetic, local-first engine for benefit valuation, unclaimed-opportunity detection and transparent next-best actions | Python, rules engine, Streamlit, pytest |
| C&Assess | A real assessment lifecycle product covering question publishing, timed attempts, autosave, scoring, analytics and organizer operations | TypeScript, Next.js, Cloudflare, D1 |
| BTC & ETH Trading Systems | Regime-aware strategy research using reinforcement learning, cross-asset signals, backtesting and risk controls | Python, Q-learning, Kalman filters, CUSUM |
AegisLoop demo · TriageLoop demo · C&Assess demo
Machine learning · LLM/VLM systems · document AI · simulation · analytics · full-stack engineering · product thinking · quant research
I start with the decision or user problem, make assumptions visible, build the smallest useful system, test the difficult paths and document what remains uncertain. I prefer explainable systems and measurable evidence over impressive-looking black boxes.
My portfolio carries the broader story—product work, consulting, strategy, leadership and case studies. This profile is the technical workspace behind it.
Abhishek Das · IIT Guwahati · work.abhishekdas@gmail.com