KRISHNENDU BISWAS

I design and run a multi-agent system on a cloud server, built on the open-source Hermes Agent. 8 specialist agents and one router take work from me over Discord and Telegram, and anything I would regret waits for my approval: sends, deletes and public pushes.

• Roster: 9 agents, one router plus 8 specialists, each with its own memory and tools.
• Channels: Discord and Telegram, mostly from my phone by voice.
• Runtime: persistent memory, 249 distinct skills and 57 recurring jobs on a schedule. 331.6 hours working with the agents since 13 Sep 2026.

Timetable, absences left per course, deadlines and friends' free slots. Schedule changes are read from college mail and confirmed by classmates. Free, no ads, works offline. An independent student project, not built or run by IIM Ranchi. Built with my AI agents.

Six short puzzle games that mirror the cognitive aptitude tests some employers use, with a timed exam mode and a coach that replays each of my moves against the best solution. Free, runs in your browser, no sign-up. An independent practice tool, not affiliated with any employer or assessment provider. Built with my AI agents.

I built an estimation platform on NestJS and React: 144 commits in 7 weeks, 7 internal applications on one platform plus 3 standalone tools.

• Wrote a 1,147-line ML estimation spec: 20 requirements, 6 regression targets over 10 engineered features.
• Harmonized QA data across 82 banking apps; reshaped 18,456 rows across 61 workbooks for upload.
• Built 2 deterministic TypeScript estimation engines with 136 unit tests, 68 checking parity with the Excel model.
• Configured AWS EC2 5-stage push-to-deploy pipeline with spec-first EARS methodology (1,322-line EARS spec, 402-line monorepo blueprint).

'Multi-Class Affect Detection in Educational Chatbots: Evidence from Imbalanced Dialogue Data' was presented at FICTA 2026 in London and accepted for publication in Springer Nature's Smart Innovation, Systems and Technologies series. Proceedings are expected in mid 2027.

• Model architecture: RoBERTa + SVMSMOTE achieving 74.19% recall on the highly imbalanced Confused class.
• Showed how to handle severe class imbalance in dialogue data from educational chatbots.

Presented a systematic review on Generative AI in Marketing at the 10th Pan-IIM World Management Conference 2025.

• Synthesized 29 Scopus-indexed research articles using the PRISMA framework.
• Evaluated adoption patterns, operational hurdles, and research frontiers for enterprise marketing.

Integrated Programme in Management at IIM Ranchi (BBA 2025, now a second-year MBA student, class of 2027), alongside the IIT Madras BS Degree in Data Science and Applications (pursuing).

• IIM Ranchi: BBA completed in 2025, second-year MBA student, class of 2027.
• IIT Madras: Diploma in Programming completed in 2026, Diploma in Data Science in progress, BS degree ongoing.
• Topper Badges in 7 of the first 14 courses of the BS program.

Competition models and applied ML work, each checked on held-out data.

• LoRA DeBERTa-v3 + RoBERTa ensemble for MCQ ranking: achieved 0.74242 MAP@3 on Kaggle public leaderboard.
• Toxic comment classification on 198K comments: F1-Macro improved from 0.6221 to 0.8293 on held-out set using a 100-trial Optuna study.
• Customer churn prediction on 90K records: tuned Random Forest achieving 0.6365 F1 score.
• RAG virtual teaching assistant: FastAPI, dense embeddings, and SQLite vector store.
• IESCP influencer-sponsor platform: full-stack Flask app.
• VC deal analytics dashboard: Power BI star schema with 7 custom DAX measures on synthetic data.
• Consumer market segmentation: K-means++ clustering on 62 respondents with silhouette score 0.540.
• Econometric modeling: 5-regressor OLS regression estimated across 28 Indian states.

Certifications in business intelligence, process improvement and supply chain simulation.

• Microsoft Power BI Data Analyst (PL-300), passed Jun 2025.
• Lean Six Sigma Green Belt: DMAIC process improvement, Grant Thornton.
• Digital Supply Chain Twins: supply chain simulation in anyLogistix, IIM Mumbai.
• Odoo x Paradox 36-hour Hackathon: top 50 of 500 teams at IIT Madras.

Pay benchmarking in industry and learning-level assessment in rural schools.

• Tata Steel, Research Analyst Intern (Apr to Jun 2025): compensation benchmarking across 20+ manufacturing firms.
• IIM Ranchi Social Impact Intern (Jan to Feb 2024): assessed 390+ rural students with Pratham and Unnat Bharat Abhiyan.

What I do when I am not building.

• Manhwa and web novels: 37,809 chapters read on Mihon (all-time favourite: Eleceed).
• Music: deep focus soundtracks and listening.
• Fintech: quantitative finance, banking systems, and financial technology.

The starting zone of the chronicle: foundational academic discipline in science and mathematics.

• Completed Class X under the ICSE curriculum.
• Secured 3rd rank across the entire school.

This page refreshes its numbers when the site rebuilds. For roles or collaborations, reach me here.