pulsedesk-os/ solo, MITcase studyrepo
PulseDesk OS
Code owns the workflow; the model makes narrow, typed judgments. 9 judgments, 5 MCP tools, 59 tests.
triagerecalldispatchdraftverify
Chandan Satwani
AI engineer, Bangalore2025–nowAI engineer, Allegion2025BTech CSE, KLE Tech, gold medal
// agent, answers only from my public projects, papers and videos// agentlive
Everything I've built, one question away.
I build grounded AI agents, fine-tune small models for the edge, and publish the research behind them.
How does PulseDesk stop prompt injection?
whybook/ solocase studyrepo
WhyBook
Chemistry tutor on Gemma, small enough to run on a laptop.
dataqloragguf
Try the live demohail-mary/ solocase studyrepo
Hail Mary Distilled
A 3B model tuned on a free GPU to admit what it doesn't know.
seedqloramergehub
Open mission control// for ai agentsin progress
Plug my work into Claude or Cursor.list_projectsask_projectget_papersget_contact
MCP server and llms.txt open with the site.Every answer is graded for grounding.An LLM judge checks replies against a public question set. Score and questions publish at launch.
// recently shipped
●PulseDesk OS v1 release●WhyBook GGUF on Hugging Face●Hail Mary model and dataset
Exploring system design for AI// research, 2025
// milestones
// chandanailabsResearch papers, explained on YouTubeAgentic AI and LLM orchestration, paper by paperGold medal, 9.6 CGPABest Boy award, 2025GDG lead, BelgaumOff the clock: space history, old machines, story RPGs
// the long line
Pick a momentEvery age builds a machine to think with.
Agents that cite their sources and admit what they do not know. This is the part of the line I work on.