# Bodham Labs

> Bodham Labs (बोधम्) is an applied AI studio in Delhi, India. It builds operations automation, private AI search and agent infrastructure for businesses in India and abroad; trains teams and students in AI, machine learning and agents, in English or Hindi; and makes Bodham, a learning platform for school students. It is a Udyam-registered MSME founded in 2026 by Prabhakar Chandra.

Key facts:

- Based in Delhi, India. Works with clients worldwide; invoices in INR or USD.
- Every client engagement starts with a small fixed-price pilot on the client's real data.
- Work is graded publicly as In use, Measured, Prototype or Exploring; work that is not measured yet is labelled as such.
- Contact: contact@bodhamlabs.com (projects), admissions@bodhamlabs.com (Bodham, students and parents), security@bodhamlabs.com (security reports).

## Work

- [All projects, graded](https://bodhamlabs.com/work): eight projects with their status and evidence.
- [RK Ops](https://bodhamlabs.com/work/rk-ops): a staff portal and WhatsApp agent that run a coaching institute's front office — absence alerts, fee reminders with UPI links, instant receipts, replies in everyday Hinglish. In daily use since October 2026; 73 of 73 automated checks passing.
- [Keepsound](https://bodhamlabs.com/work#keepsound): an on-device voice memory with hybrid search (keywords, meaning, dates, names) that runs fully on an Android phone.
- [Nexus RAG](https://bodhamlabs.com/research/agent-memory): a local long-term memory server for AI coding agents over the Model Context Protocol (MCP); used daily in our own work from March to May 2026.

## Services

- [Operations automation](https://bodhamlabs.com/services#automation): reminders, receipts, follow-ups and reports; client deployments use the official WhatsApp Business Platform.
- [Private AI search](https://bodhamlabs.com/services#search): answers from a client's own documents, with sources; can run on the client's own machines or on a phone.
- [Agent infrastructure](https://bodhamlabs.com/services#agents): memory and tool servers for AI agents over MCP.
- [Hindi & Hinglish AI](https://bodhamlabs.com/services#hinglish): speech recognition, search and replies for Hindi, English and mixed speech.
- [Data extraction](https://bodhamlabs.com/services#data): structured data from web pages and files.

## Research

Unproven work is labelled Exploring; dead ends are written up too.

- [All research](https://bodhamlabs.com/research): what we are testing and where it has led.
- [Long-term memory for AI agents](https://bodhamlabs.com/research/agent-memory): a local MCP memory server; why the language model was taken out of the query path; GLiNER instead of a small LLM for entity extraction.
- [Keeping autonomous agents from breaking themselves](https://bodhamlabs.com/research/agent-reliability): a 45-hour agent outage and the guardrails written after it.
- [State space models for long data streams](https://bodhamlabs.com/research/state-space-models): a Mamba (S6) run whose hidden state exploded on un-normalised inputs; the checklist we use now.
- [Real-time ML pipelines on modest hardware](https://bodhamlabs.com/research/realtime-ml-pipelines): a Python-to-C++ shared-memory pipeline at 100 Hz for 24 hours with zero crashes; latency spikes cut from 2,075 ms to 111 ms.
- [Context that survives between AI coding sessions](https://bodhamlabs.com/research/agent-context-tooling): NeoGravity, published on npm as @vibcoder/neogravity.
- [Making businesses readable to AI answer engines](https://bodhamlabs.com/research/answer-engine-optimisation): a store-catalogue-to-Markdown prototype for AI crawlers.
- [Faithful Hinglish translation with LLMs](https://bodhamlabs.com/research/hinglish-translation): the style rules that kept an LLM translation faithful.

## Training

- [Training programmes](https://bodhamlabs.com/training): "AI at work" for teams, "Build with LLMs & agents" for developers, "Python & ML from zero" for students, and programmes for colleges. Taught in English or Hindi.

## Notes

- [Vector search on an Android phone: 365,000 chunks in 7 ms](https://bodhamlabs.com/notes/vector-search-on-android): exact int8 search over 365,000 vectors took 20.2 ms on one thread and 6.9 ms on four on a Redmi Note 14 Pro+ 5G — 10–18× faster than estimated.
- [Why our agent memory stopped waiting for a language model](https://bodhamlabs.com/notes/language-model-out-of-search-path)
- [Our autonomous agent went dark for 45 hours. These are the rules we wrote after.](https://bodhamlabs.com/notes/agent-45-hour-blackout)
- [Our Mamba model's hidden state reached 10^15. Normalise your inputs.](https://bodhamlabs.com/notes/mamba-hidden-state-exploded)
- [2-second stalls from one synchronous disk write](https://bodhamlabs.com/notes/sync-write-latency-spikes)
- [The pipeline ran 24 hours without a crash. The model still failed.](https://bodhamlabs.com/notes/pipeline-worked-model-failed)
- [Translating into Hinglish with an LLM: the rules that stopped the drift](https://bodhamlabs.com/notes/hinglish-translation-rules)
- [From web developer to an applied-AI studio: how Bodham Labs started](https://bodhamlabs.com/notes/from-web-developer-to-applied-ai)
- [All notes](https://bodhamlabs.com/notes): benchmarks, field notes and guides.

## Company

- [About Bodham Labs](https://bodhamlabs.com/about): facts, founder, and the meaning of the name.
- [Contact](https://bodhamlabs.com/contact): email, or a free 30-minute call from anywhere.
- [Careers](https://bodhamlabs.com/careers): apprenticeships for BCA and engineering students — apprentice developer, automation operator, teaching assistant.
- [Bodham](https://bodhamlabs.com/bodham): study and practice for school students in Hindi and English (launching soon).

## Optional

- [Notes RSS feed](https://bodhamlabs.com/notes/rss.xml)
- [Sitemap](https://bodhamlabs.com/sitemap.xml)