of every work week is lost hunting for information
Cottrill Research
Taksha
How does a company
How does a company
find what it knows?
Knowledge management for organizations: keeping the judgement, not just the files.
built with Aarya
home
01 · the problem
Every company forgets.
This one found me before I went looking. Day one of my Info Edge internship (the same summer as case 03): a laptop, a Slack account, and a polite avalanche.
The PRD was in Confluence. Decisions in Slack. Tickets in Jira. Numbers in Sheets. The context was in someone's head, and that someone was in meetings until Thursday. A year later, Aarya and I compared internship notes and found the same wound in both.
day 01 · fresh & clueless
aarya · same summer
Where does what-we-know actually live?
new joinee
“where's the latest PRD?”
slack
“check notion. or the deck. maybe both.”
same joinee
“found three versions. which one is real?”
the senior
“just hop on a call, it's faster if I explain.”
Many tools.
Zero memory.
Zero memory.
Every answer exists somewhere. Nobody knows where, and the person who does is the bottleneck.
skipped on mobileA pinned scene of six tools and four overheard quotes colliding as you scroll, it needs the room to feel like chaos. The short version: every answer exists somewhere, nobody knows where, and the person who does is the bottleneck.
And this isn't an intern problem. It's an industry-sized tax that shows up in every study we could find:
Cottrill Research
won back by Duolingo by fixing internal search
Work AIOnly 0% of employees say onboarding actually works.
Gallup
call KM criticalfeel ready to act
Deloitte
When experts leave, decades of tacit insight walk out too
Businesswire
Companies don't lose files. They lose judgement: the why, the almost, the “we tried that in 2022”.
Explicit Knowledge: Knowing what
Structured. Easy to document. Easy to share.
Policies, manuals, tutorials, databases, memos
Implicit Knowledge: Knowing how
Applied through practice and context.
Learnings from shadowing, repeated tasks, internalized logic
Tacit Knowledge: Knowing why & when
Intuitive. Experience-driven. Hard to codify.
Judgement, culture, team memory, mental models, unspoken know-how
And tacit is the toughest to capture, the layer no document, wiki or handover call ever reaches.
the uncomfortable truth
scoping the thesis · why india?
The most valuable knowledge in a company has never been written down. Every tool waits for someone to write it down.
We anchored the thesis in Indian knowledge companies, not for convenience, but because the wound is deepest here:
Scale
5M+ knowledge workers
And a SaaS market racing to $50B. Every one of those companies is compounding the same forgetting.
Attrition
IT attrition ≈ 15%
Every exit drains hard-won know-how. At Indian churn rates, memory loss is a quarterly event.
AI-readiness
92% use AI at work
Yet 54% of leaders have no knowledge-management plan. Appetite without infrastructure.
Culture
47% feel job-insecure
Insecurity breeds knowledge hoarding, sharing what you know can feel like training your replacement.
the problem we arrived at
The gap isn't retrieval, it's capture. The knowledge that runs a company never gets written down, and nobody is rewarded for sharing it.
so, the brief we set ourselves
Give organizations a memory that compounds: capture what teams learn, keep it alive, and serve it back in the flow of work.
A brief isn't an answer. Desk research could name the problem, not why people stay quiet. So we stopped reading and went to ask.
we couldn't be the only ones who noticed, so…
How are companies solving this today?
02 · research
How teams actually remember
They are, and that was worth knowing, a whole category already exists. The problem was validated before we drew a thing. What none of it closed was the gap we kept falling into.
primary · people
We sat with the people who lose the context
- One-to-one interviews with ICs, seniors and managers
- Screen-share tasks, “show me how you'd find X”
- Think-aloud inquiry during real work, then thematic analysis
secondary · desk
Then checked it against what the field already knew
- Academic KM literature and industry reports
- Competitive benchmarking across the KM landscape
- Rollout post-mortems, the ones that stuck and the ones that didn't
“Even if the docs are there, I don't know where to start.”
new joinee · product
“Honestly? I just ask the senior sitting next to me. It's faster.”
engineer · 8 months in
“Documentation is nobody's job. So it becomes everybody's regret.”
engineering manager
“When our analytics lead left, half the funnel context left with her.”
product manager
01
Capture is the bottleneck
Writing things down is extra work with zero reward. Knowledge gets shared in meetings and DMs, the places no wiki ever sees.
02
Retrieval is role-blind
The same wiki greets an intern and a CTO identically. Nobody gets knowledge shaped to their role, project or level of context.
03
Onboarding is apprenticeship
New joiners learn by shadowing seniors. It works beautifully, and scales terribly, taxing the exact people with the least time.
The industry is racing to search documents. Almost nobody is capturing what never becomes one.
We scored the tools companies actually buy against our workflow maps. Strong products, all of them, all stopping at the same wall:
Glean
enterprise searchThe reference for retrieval: permission-aware search across 100+ tools.
Searches what's been written, never what was said or decided.
Guru
verified wikiVerification and staleness alerts keep answers trustworthy.
Runs on manual curation, and nobody has a reason to curate.
Notion AI
workspace Q&AExcellent answers, if your team already lives in Notion.
Workspace-locked. Slack threads and meetings stay invisible.
Microsoft Viva Topics
big-tech attemptThe most ambitious capture attempt: auto-topics across the M365 graph.
Retired in 2025. Even Microsoft couldn't make passive extraction stick.
capability
GleanGuruNotion AITaksha
AI retrieval across toolsone query, every system✓~~✓
Captures tacit knowledge at the sourcethe "why", not just the doc✕✕✕✓
Surfaces who to ask, not just whatownership as a first-class object✕✕✕✓
Incentives to contributewriting back is rewarded, not chores✕✕✕✓
Project & decision memorythe trail behind a shipped call✕~~✓
Flags stale or contradicting knowledgememory that ages honestly✕~✕✓
✓ native~ partial✕ absent
the differentiator
Better search wins a feature war. Whoever solves capture + motivation owns the category, because they own the supply of knowledge itself.
design decision
Everyone was fighting over retrieval, so we went after supply instead: get knowledge out of heads without asking anyone to write documentation, and make sharing feel like winning rather than training your replacement.
03 · the process
Designing a memory, not an app
A system this broad dies the moment it becomes “a bunch of screens”. So we built it in a fixed order, and refused to draw a single screen until the first three steps held.
- Whotwo seats at opposite ends of the knowledge pipe
- Rulesfour tenets that outrank any screen
- Systemone diagram everything traces back to
- Surfacewhere that system had to live
- Namewhat we were actually building
- Screensmaps → objects → lo-fi → hi-fi
Every flow got walked in two pairs of shoes, two seats at the same company, at opposite ends of the knowledge pipe:
Aditya · the new joinee
Week one, product analytics. Knows nothing about how this team works. Consumes knowledge; can't yet tell signal from noise.
knowledge sink
Yash · the senior
Five years in, carries half the team's context in his head. Everyone's favourite bottleneck. (Yes, we named him after me. Method acting.)
knowledge sourceWith those two on the wall, each finding hardened into a rule that outranked any screen:
tenet 01
Capture at the source
Meetings, PRs, chats: knowledge gets caught where it's created, not where someone remembers to file it.
tenet 02
Live in the flow of work
No “knowledge portal” anyone has to remember to visit. Taksha meets you where work already starts.
tenet 03
Role decides the view
A new joinee and a senior never need the same answer. Context shapes everything served.
tenet 04
Sharing must pay
If contribution isn't visibly rewarded, the 47% who hoard will keep hoarding. Rationally.
Four rules, four pillars. Everything below the pillars is plumbing the user never sees; everything above it is a face on one of the four:
TakshaRecordingknowledge
Retrievingknowledge
Role-specificonboarding
Automationagents
🧠 Collective Company Brain + RAG framework + NLP
your datasources
17 connectors at thesis scope, read-only, permission-aware
17 connectors at thesis scope, read-only, permission-awareyour cloud
knowledge never leaves your walls
knowledge never leaves your wallsEvery screen in chapter 04 is one of these pillars wearing a face. That was the test I kept applying: if a screen can't point to its pillar, it's decoration.
guide checkpoint · mid-review
step 04 · where it lives
“The system diagram is the thesis. The screens are just proof it can exist.”
Our guide, Prof. Dr. Pratul C. Kalita, signed off on the direction at mid-review with one condition, every feature demo had to trace back to a pillar and a research finding. It became our editing knife.
the obvious move
Another web app
One more tab to forget, one more login to skip. Knowledge tools die of “I'll check it later”.
our call
The new-tab surface
Taksha lives where work already begins: the browser's new tab. Zero habit change, every search, meeting and to-do passes through it anyway.
तक्ष
Takṣa
Takshashila (Taxila) is the oldest recorded university on the subcontinent: a place people travelled to because knowledge was kept, not hoarded. That was the ambition, so the name came from its root, takṣa: to shape raw material into something usable.
Which is the product's job. Raw material in, threads, tickets, half-decisions; something a new joinee can pick up out. Not a vault. A workshop.
Only then the drawing, and by then it was almost mechanical: workflow maps per persona → an object model (Playbooks, Story Cards, Sparks) → lo-fi → hi-fi. Weekly guide reviews and hallway tests on classmates who'd just come back from product internships kept each pass honest.
04 · the design
Meet Taksha
Named after Taxila, the ancient university where knowledge compounded for centuries. The demo runs as a story in two seats: Aditya's first week, then the senior's worst Tuesday.
the surface
A new tab that opens as a briefing.
- 01 today's meetings, tickets and gaps
- 02 one question box, everything behind it
- 03 zero new habits, it's the tab you already open
It opens like a briefing, not a portal: your meetings, your tickets, one flagged gap, one question, “what are we curious about today?”. Everything else hangs off that sentence.
seat 01 · pillar: role-specific onboarding
Aditya's first week
No orientation PDF. Taksha builds his first week for him. Four moments, in order:
01
day 01 · 09:04
A dashboard that knows him
His onboarding course, his meetings, his team's files, all assembled from role + project context before he asks for anything.
02
day 01 · 09:11
A course curated by AI
“Welcome Aboard: Product Analytics Starter”, generated for his exact role, then approved by a human lead.
03
day 01 · 09:26
A syllabus with receipts
12 lessons, 4 self-paced hours, a mentor one Slack click away. Every lesson traceable to real team files.
04
day 01 · 11:40
Progress that pays
Sparks and badges land from the first module. Momentum is designed in, not hoped for.
design decision
learning by errand
No receipts, no render. Every generated artefact has to name its sources, and “from 14 files” is a link, not a reassurance. Testers stopped reading AI output as a rumour the moment they could click it.
Instead of a quiz about the tool, Aditya gets a real errand with a 30-minute timer:
05
day 03 · 14:02
The quest
“Your PM needs daily active users by platform, in 30 minutes.” Exactly the panic he'll face for real in week two.
06
day 03 · 14:09
The intended shortcut
Smart Search surfaces the team's vetted SQL snippet, endorsed by 12 peers, instead of twenty minutes of Slack archaeology.
07
day 03 · 14:21
The payoff
Right query, +10 Sparks, and the habit that matters: search the memory first, interrupt a human second.
Past week one, the same engine becomes a university built from the company's own memory.
Same engine, three jobs: ramping up new joinees, moving people between squads without the three-week context beg, and turning exit interviews into capture instead of goodbyes.
smart search, everywhere
seat 02 · pillars: recording + automation
The senior's worst Tuesday
Back from two weeks of leave, coffee still brewing, and this lands in his inbox:
a report. on a project he missed. by 4pm.
The old version of this: two hours of Slack scrollback, three “quick calls”, a deck assembled from fear. The Taksha version is a walk through the project's memory:
pillar: recording · playbooks
playbooks as a living system, not a graveyard of docs
Daily work
→
Project playbooks
→
Departmental playbooks
→
New projects
+ retrieval & automation
+ retrieval & automation
Capture happens where knowledge is born, mid-meeting, not in a doc nobody writes.
design decision
project memory · the timeline
Viva Topics scraped quietly and died quietly. So capture asks for one human yes, and it asks in the second someone is already thinking “can we save that somewhere?”
Every project keeps a queryable spine of what happened while he was away:
04
tue · 10:12
The project space
Chats, quick links, calendar and mentions for “New user onboarding 2.0”. One address for the whole effort.
05
tue · 10:19
The timeline view
Kickoff → feedback → direction change, as nodes with attached files, owners and Slack threads.
06
tue · 10:31
Zoom into any node
The mid-milestone review: demos shown, feedback recorded, and the exact quote that changed direction, timestamped.
07
tue · 10:44
Approve what enters memory
Pending playbook entries wait for a human yes. The memory stays curated, not scraped.
Then the memory starts working for you: agents that turn retrieved knowledge into finished chores.
His 4pm report? BriefBot reads the PRs, the Jira tickets and the team's tone of voice, and returns a Slack-ready brief. Draft compiled in ≈ 25 seconds; the manual version took him 40 minutes last quarter.
If sharing doesn't pay in something visible, the 47% who hoard keep hoarding, rationally. So: a Spark for the act, a constellation for the pattern, a galaxy for the quarter.
01
Sparks
Every captured insight, endorsed answer or finished quest mints a Spark, one unit of “I made the memory richer.”
1 act · 1 spark
02
Constellations
Endorsed sparks link into constellations: visible patterns of what a team actually knows.
many sparks · 1 pattern
03
Galaxies
Over quarters, constellations settle into galaxies, a searchable sky of company memory.
a quarter · 1 sky
the guardrail
Sparks reward usefulness, scored by peer endorsement, never volume. Pay for volume and you've designed a spam machine with badges.
All of it came together as a role-played prototype, Aditya's week and the senior's Tuesday, demoed end to end.
05 · learnings
What the thesis taught us
Taksha is a thesis, not a shipped product, so the honest scoreboard is what it proved, what it taught, and what we'd bet on next.
the room's verdict
as a designer
The capture-to-incentive loop landed as the strongest idea in the final jury.
Prof. Kalita backed the direction at every review; his consistent push was rigor, trace every screen to a finding. The jury's read matched our research bet: retrieval is everywhere, supply is the open problem.01
Systems before screens
The four-pillar diagram did more editing than any critique. When a screen couldn't name its pillar, it died, and the product got sharper each time.
02
AI needs receipts
Trust in AI output is a design problem, not a model problem. “Generated from 14 files”, clickable, changed how testers treated every answer.
03
Incentives are design material
The sparks economy needed balancing like a game: reward usefulness, not volume, or you've designed a spam machine with badges.
04
Role-play beats feature tours
Demoing through Aditya and the senior forced every feature to earn a place in someone's day. It's how this case study is written, too.
01
The gap is supply-side
Glean, Guru and Notion fight over retrieval. Capture + motivation is an open category, and the moat compounds, because memory does.
02
Onboarding is the wedge
“Cut ramp-up from 3 months to 3 weeks” is a number a CFO can love. Land there, expand into the full memory.
Beyond the thesis: pilot with one real team and watch three numbers, time to a new joiner's first contribution, the endorsement rate on captured sparks, and whether the senior's interruption load drops. If those move, the rest follows.
next
Auto-generated case studies
Workstreams turned into branded, publishable stories.
next
Knowledge marketplace
Monetizing reusable agents and playbooks across orgs.
next
Voice-first interface
Hands-free capture and retrieval, in the hallway where knowledge actually moves.
credits
A two-person bachelor thesis, built with Aarya Ghodke.
Equal partners on every pillar, so this telling walks my half of it. Guided by Prof. Dr. Pratul C. Kalita, and named for a university that kept knowledge alive for a thousand years, which felt like the right bar for a company's memory.